ABSTRACT AN EXPERIMENTAL AND THEORETICAL INVESTIGATION INTO NONFLAME ATOMIZATION By Scott Roy Goode The widespread application of nonflame atomization to routine atomic absorption (AA) or atomic fluorescence (AF) analyses is a relatively recent phenomenon. Even though nonflame AA or AF analysis is not as precise or as easy to perform as is flame AA or AF analysis, the high sensitivity achieved by nonflame atomization methods makes it a valuable tool. An experimental and theoretical investigation was performed to see if the major sources of imprecision could first be isolated and then minimized. The design of a completely automated nonflame atomic fluores- cence spectrometer is described. The system was used to investigate the phenomena occurring at a very simple nonflame atomizer, an elec- trically heated platinum loop atomizer. The extent of automation made it feasible to perform tens of thousands of analyses in order to characterize fully the effects of the instrumental parameters on the final readout. A study of the events occurring at the atomizer is described. The fundamental study of atomization and the experimental study of Scott Roy Goode the effects of the instrumental parameters on the readout were used to fully optimize the nonflame spectrometer. Data taken by the fully optimized system are presented and critically analyzed. The results of the study of atomization were used to design a new spectrometer which uses a braid of graphite fibers as an atomizer. This new atomizer is characterized and analytical data are presented to demonstrate the overall utility of the graphite braid atomizer. AN EXPERIMENTAL AND THEORETICAL INVESTIGATION INTO NONFLAME ATOMIZATION By Scott Roy Goode A DISSERTATION Submitted to Michigan State University in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY Department of Chemistry 1974 Q 575' LN To my Mother and Father They understand ii ACKNOWLEDGMENTS The author would like to express his sincere and deep gratitude to all those who helped him in his research. The primary moving force was supplied by Dr. Stanley R. Crouch, whose confidence and optimism helped minimize the dry spells, and whose ability and perspective helped maximize the productive periods. One can never reach the source flicker limit when the source is unwavering. Thanks also go to Dr. Christie G. Enke, my diligent second reader, and to Dr. Fred Horne for their helpful discussions. Michigan State University is to be thanked for their support in the form of assistantships, as well as their support from an able staff. Among the people that can be singled out are Ron Haas and his people in the Electronics Shop, and Russ Geyer and his crew of machinists. Thanks also go to the secretarial staff for their help. The members of my research group, particularly Dave Rothman, deserve special mention. Not only would they listen to my hypotheses without laughing, but they would suggest experiments to prove them. My only regret is that they did not suggest the incredibly complex experiments that could possibly verify my theories, but always thought of some fundamental, simple experiment that disproved my work. I would also like to thank Lee and Carol Thomas who opened their hearts and their home, and taught me some basic Humanities. Last, but certainly not least, I would like to thank Rose and Tom McCowen who provided typing, artwork, technical aid, and a bottomless stewpot. Without their assistance, this work would have shrivelled due to malnutrition. TABLE OF CONTENTS Page LIST OF TABLES .......................... viii LIST OF FIGURES ......................... ix 1. INTRODUCTION ....................... l A. Flame Atomization .................. 3 B. Nonflame Atomization ................ 5 C. Filament Atomizers ................. 7 II. HISTORICAL ........................ 8 A. Furnace Atomizers .................. 9 l. The L'vov Furnace ............... 9 2. The Massman Furnace .............. l2 3. The Noodriff Furnace ............. 13 4. Other Furnaces ................. 14 B. Filament Atomizers ................. 20 l. The Nest Atomizer ............... 20 2. Commercial Graphite Filament Atomizers ..... 27 3. Other Graphite Filament Atomizers ....... 29 4. Metal Filament Atomizers ------------ 3l C. Conclusions ..................... 34 III. THEORETICAL DESCRIPTION OF ATOMIC FLUORESCENCE FROM A NONFLAME ATOMIZER ................ 35 A. The Experimental Observation of Atomic Fluorescence .................... 36 B. Expressions Describing the Line Radiance of Atomic Fluorescence ................ 37 1. Atomic Absorption with a Continuum Source . . . 37 2. Atomic Absorption with a Line Source ...... 40 3. Radiance Expressions forAtomic Fluorescence . . 4T iv C. Characterization of the Atom Population of a Nonflame Atomizer ............... 45 l. Time Dependence of Atom Population . . . . 45 2. Time integrated Atom Population Expressions ............... 48 D. Time Dependence of Temperature ......... 50 1. Heat Sources and Heat Sinks in a Filament . ............... 50 2. Zero Order Perturbation Terms ....... 52 3. First Order Perturbation Terms ...... 57 E. Modified Estimates of Atom Population ..... 59 IV. INSTRUMENTATION ................... 64 A. Design Criteria ................ 65 l. The Atomizer ............... 65 2. The Spectrometer ............. 67 8. Sequence of Events ............... 70 C. Sampling to the Atomizer ............ 7l D. Electrical Control of Atomizer Temperature . . . 74 E. Optical Design ................. 79 1. Primary Excitation Sources for Atomic Fluorescence ........... 79 2. The Atomizer ............... 8l F. Photocurrent Integration ............ 86 l. Analog Integration ............ 86 2. Digital Integration ............ 91 3. Photon Counting .............. 93 V. OPTIMIZATION OF INSTRUMENTAL PARAMETERS ....... 98 A. Optimization of Signal-to-Noise Ratio . . . . 98 8. Experimental ................. 100 l. Instrumentation ............. lOO 2. Procedure ................ lO4 C. Experimental Variables ............ 104 l. Sample Size ............... l06 vi 2. Sheath Gas Flow Rate .......... 107 3. Power Applied to Atomizer ........ 110 4. Integration Parameters ......... 112 D. Interaction of Variables ........... 114 E. Analytical Results .............. 119 1. Stability ................ 119 2. Calibration Plot ............ 119 VI. CHARACTERIZATION OF THE ATOMIZATION PROCESS . . . . 128 A. Characterization of the Platinum Filament Atomizer .................. 128 1. Evaporation of the Filament Material . . 128 2. Changes in the Physicochemical Properties of the Platinum Filament Atomizer ................ 130 B. Events at the Atomizer Surface ........ 132 l. Desolvation ............... 132 2. Atomization ............... 133 3. Time Resolution of Events Prior to Atomization .............. 134 4. Post Atomization Events ......... 137 VII. APPLICATION OF EXPERIMENTAL RESULTS ........ 144 A. Optical Considerations ............ 145 1. The Monochromator ............ 145 2. Nondispersive Optical Systems ...... 154 B. Photocurrent Integration ........... 155 C. The Sampling System ............. 156 D. Temperature Programming ........... 156 E. Analytical Applications ........... 160 1. Optimization for Selectivity ...... 160 2. Calibration Plot ............ 161 F. The Graphite Braid Atomizer ......... 164 1. Advantages ............... 164 2. Optimization of the Graphite Braid Atomizer ............... 166 vii 3. Analytical Applications ........ 168 VIII SUMMARY ...................... 173 LIST OF REFERENCES ................... 175 Table 01-th LIST OF TABLES Comparison of Methods of Peak Integration . Optimum Spectrometer Parameters Interaction of Instrumental Parameters . Optimum Platinum Loop Atomizer Parameters . Spatial and Temporal Atom Distribution as a Function of Time. Experimental Conditions for the Determination of Stray Radiant Energy Matrix Effects of Platinum Loop Atomization The Signal at the Limit of Detection. Effects of Atomization Temperature on Cadmium in Several Matrices . . . . . . . . . . viii Page 93 106 117 118 139 153 162 170 172 Figure 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. LIST OF FIGURES The Atomic Vapor Cell . Growth Curves for Atomic Fluorescence Dependence of Atom Concentration on Residence Time in Observation Cell . Dependence of Atom Concentration on Atomization Time Comparison of Theoretical Prediction of Atom Concentration with Experiment Comparison of Theoretical Predictions of L'vov (86) and this Work . . . . . . . Sample Delivery System Block Diagram of Triac Power Control Circuit Circuit Diagram of Triac Power Control Circuit Atomizer Assembly Optical Diagram of the Atomizer Assembly Fixed Time Integration Digitization Circuit Photon Counting Circuit Photon Count Rate as a Function of Direct Current Block Diagram of Automated Nonflame Spectrometer . Circuit Diagram of Spectrometer Control System Dependence of Precision on Sheath Gas Flow Rate Dependence of Integrated Signal on Power Applied to the Atomizer . . . . ix Page 38 44 47 49 61 62 73 77 78 82 84 87 89 94 97 101 103 109 111 Figure Page 20. Dependence of Precision on Power Applied to the Atomizer . . . . . . . . . . . . . . . . 113 21. Dependence of Precision on Integrator Parameters . . . 115 22. Long Term Stability of Nonflame Spectrometer . . . . 120 23. Integrated Fluorescence as a Function of Cadmium Concentration Under Fully Optimized Conditions . . . 122 24. Peak Atomic Fluorescence as a Function of Cadmium Concentration Under Fully Optimized Conditions . . . 124 25. The Signal at the Limit of Detection . . . . . . . 125 26. Photon Count Rate as a Function of Cadmium Concentration. 126 27. Time Resolution of Events at the Platinum L00p Atomizer . 127 28. Peak Atom Population Profiles . . . . . . . . . 143 29. The Slit Transfer Function. . . . . . . . . . . 147 30. Comparison of the Varying Slit width and Limiting Absorbance Method of Determining SRE . . . . . . 152 31. Dependence of Integrated Fluorescence on Ashing Temperature . . . . . . . . . . . . . . . 159 32. Ca1ibration Plot for Cadmium in Several Matrices Obtained with Temperature Programmed Atomization . . 163 33. Dependence of Integrated Fluorescence on Atomization Temperature of GBA . . . . . . . . . . . . 167 34. Calibration Plot; GBA Fully Optimized . . . . . . . 169 I. INTRODUCTION One of the largest problems facing the analytical chemist is in the nature of sample treatment prior to analysis. The ideal elemental analysis is independent of the history of the sample. The results of a cadmium ana1ysis, for example, should not be altered if the cadmium is in an aqueous solution, organic solvent, or protein matrix. One method of eliminating matrix effects is to break the compound into its constituent atoms, and then to per- form the analytical measurement on the atomic vapor. The atoms, once produced, exhibit no memory effect. It is precisely this advantage which makes atomic spectrometry such a powerful analytical technique. The three atomic spectrometric techniques of atomic emission, atomic absorption, and atomic fluorescence all require the produc- tion of an atomic vapor from the analytical sample. The average ana1ytica1 sample must undergo a series of steps before it is atom- ized, and these steps are the same in all three atomic spectrometric techniques. First, the solvent must be stripped from the analyte (assuming it is in a solution) to form crystals of a salt of the analyte. Second, the crystals must be vaporized, and third, the mo1ecu1ar vapor must be broken into its constituent atoms. If the atomic vapor is sent into a heated cell, such as a flame, some of the atoms will be excited, in accordance with the Boltzmann distribution. When the radiational deactivation of a thermally excited atom is observed, the phenomenonis;called atomic emission. The "intensity" of the emission signal is proportional to the ex- cited state population, which in turn is proportional to the ground state population. The emission signal, or the integrated line radiance, to be more exact, is a linear function of concentration at low concentrations and is proportional to the square root of concentration at high concentrations. In practice, the calibra- tion plots for atomic emission are linear over three to four orders of magnitude. Atomic absorption spectrometry (AAS) and atomic fluorescence spectrometry (AFS) both require an external radiation source to excite the ground state atoms. In AAS the attenuation of the ex- ternal source radiation is measured, while in AFS the intensity of radiation emitted after radiational activation is observed. The fluorescence radiance is proportional to the radiance of the source, and in theory, AFS is never less sensitive than AAS. Only the lack of intense sources has prevented AFS from becoming widely used in analytical spectrometry. Obviously, for any of the three atomic spectrometric methods to work, the number of analyte atoms produced must be proportional to the number of analyte atoms in the sample. For the best sensi- tivity, the atomization efficiency should be 100%, and should not be dependent on matrix. The sample matrix usually does not affect the measurement performed on the atomic vapor. When the sample is broken into its constituent atoms, only a fraction of the total analyte will appear as neutral, ground state atoms within the area of observation. The fraction atomized is generally dependent on matrix. A. Flame Atomization The most common method of atomization is to spray a liquid sample into a flame. The flame first evaporates the solvent, leav- ing particles of the analyte salt. The particles are vaporized into a molecular gas, and the molecules are finally atomized. The atomization process is thus dependent on the type and chemical properties of the molecules formed. Aluminum chloride particles for example, are more easily atomized than aluminum oxide particles. Flames have been successfully used as atomizers for many years (1). The use of flames as atomizers was inherited from the tech- nique of flame emission spectroscopy where the flame was used both to atomize the sample and to excite the atoms. Even when used only as an atom reservoir for AAS and AFS, the chemical combustion flame has many advantages. 1. Flames are convenient, reliable, and reasonably safe. 2. The atom population is due to the vaporization and atom- ization of large numbers of particles. This leads to high precision due to the effective averaging of many individual particles. 3. A variety of flames are available with chemical environ- ments ranging from strongly oxidizing to strongly reduc- ing. Flames have temperatures from 520 K to 5900 K (1). The wide choice of flames allows the spectroscopist to choose appropriate conditions to atomize different samples. Flames, however, also have several disadvantages which limit their applicability. l. The flame is a dynamic, poorly understood system consist- ing of a mixture of violent chemical reactions. The re- actants, products, and their intermediates may all be excited and produce intense background emission. The analytical signal may be lost in this large background. 2. Flames generally cannot atomize solid samples. In fact, most flames require about 1 m1 of liquid for an analysis, so that micro-samples cannot be analyzed. 3. The expansion of the flame gases dilutes the atomic vapor. This limits the maximum attainable atom concentration. 4. Precise control cannot be exercized over the chemical environment. Oxygen supported flames will always con- tain some oxygen, no matter how rich the mixture. The oxygen may react with the analyte to form refractory oxides which are very difficult to atomize. The flame products, in addition to reacting with the analyte, may also undergo collision with the excited state atoms and provide a path for radiationless deactivation. In general, the flames which are the best atomizers are also the best quenchers, so atomic fluorescence is seldom attempted in these flames. The first two disadvantages are both practical in nature. It is possible to choose a flame which does not exhibit intense emission at the wavelength of interest, and is conceivably possible to design a flame system for small samples and for solids. The latter two disadvantages are fundamental in nature. Expansion of flame gases will always dilute the analyte atoms, and full con- trol over the chemical environment around the analyte cannot be achieved. The disadvantages of flame atomization are manifested in detection limits which are far worse than predicted by theory (2), and in the need for relatively large sample volumes. The entire information contained in an analytical sample cannot be obtained unless the analysis technique is sufficiently sensitive. In order to improve the quality of the analyses, many researchers have in- vestigated nonflame atomization as an alternative to flame atomiza- tion. 8. Nonflame Atomization The simplest nonflame atomizers consist of a heating element, such as a graphite filament, which atomizes the analytical sample. The nonflame atomizers have several advantages over flame atomizers. l. The sample transport is nearly 100% efficient. This is in contrast to flame atomizers where, at best, only 15% of the sample is even presented to the flame (3). 2. The atomization process is also more efficient in a non- flame atomizer than in a flame. For certain elements, the efficiency has been estimated at 100% (4). This produces a higher concentration of atomic vapor than normally found in a flame (5). 3. Control over the chemical environment can be maintained by surrounding the atomizer with the appropriate atmosphere. Argon might be selected if atomic fluorescence is desired, because argon has a low quenching cross-section. The atom- ization temperature is also controlled electrically and can be optimized for individual elements. The only background emission is the black-body type continuum emitted by the atomizer. The background emission is separated from the analytical signal by either modulation techniques as are used in flame spectrometry, or by viewing the atomic vapor at a spatial location away from the atomizer. Although nonflame atomizers have several fundamental advantages when compared to flame atomizers, the advantages are not fully realized in practice. The relative imprecision in placing a small sample on the atomizer may limit the overall precision of the analy- sis. The complexity and lengthy analysis procedure impose practical limits on the apparent widespread applicability of nonflame atomiza- tion (6)- C. Filament Atomizers In order to design an atomizer which incorporates most of the advantages of both flame and nonflame atomization, with few of the disadvantages, a fundamental study was undertaken to describe the processes occurring during atomization from a platinum loop atomizer and the dependence of the atomization process on the various experi- mental variables. The variables were found to be dependent on each other, and the fundamental phenomena underlying the dependence of the atomization and each of the variables are described. From a knowledge of how the variables effect the atomization and how they affect each other, optimum conditions could be rationally chosen. Analytical data are presented for a nonflame atomizer in which parameters have been fully Optimized. The fundamental study of atomization from a heated wire fila- ment, described in this thesis, and studies conducted by Mr. Akbar Montaser (7) in these laboratories led Mr. Montaser to propose a new filament atomizer with superior characteristics. The new atomizer, the graphite braid atomizer, was used in conjunction with a new atomic fluorescence spectrometer, designed specifically for use with nonflame atomizers. The analytical results are quite promising, and the graphite braid atomizer (8) exhibits the potential to be of widespread use in nonflame atomic fluorescence spectrometry. II. HISTORICAL The applications of nonflame atomizers in atomic absorption and atomic fluorescence spectrometry have been reviewed by Kirk- bright (9) and more recently by Winefordner and Vickers (10). The technique of nonflame atomization actually pre-dates flame atomiza- tion for both AA (11) and AF spectrometry (12). The first analytical application of nonflame atomization was reported by L'vov (13) in 1959,and only a few papers on nonflame atomization were published prior to 1969. Since the introduction of commercially available nonflame atomizers in 1971, the number of publications has grown so rapidly that only the most important advances will be reviewed in this section. The nonflame atomizers most widely applicable to AA and AF spectrometry are "thermal" atomizers in which the sample is atomized by the application of heat. Several other mechanisms have been investigated including a long-path stabilized DC arc (14), a pulsed arc atomizer (15), RF plasmas (16, 17) and hollow cathode excitation (18). A promising technique is that of laser sampling (19). A not so promising technique is the "explosion" atomizer (20) where the sample is mixed with a solid prepellant powder and exploded in the light beam of an atomic absorption spectrometer. Only one publica- tion exists concerning this technique. Thermal atomizers fall into two main classes, based on design. Furnace-type atomizers confine the atomic vapor within the atomizer, whereas filament-type atomizers allow the vapor to diffuse. Each type has inherent advantages and disadvantages which are discussed in the sections to follow. A. Furnace Atomizers 1. The L'vov Furnace The first furnace used as a source of neutral atoms was the King furnace (21). The sample was placed in a graphite tube furnace, the furnace was evacuated, and then heated. L'vov modified the King furnace for use as a nonflame atomizer for AA spectrometry (13). The L'vov furnace itself consists of a tube of graphite, 30 to 50 mm long and 2 to 4 mm i.d. The analytical sample is placed on a graphite electrode, and the electrode is inserted into the furnace to form a "T"-shaped assembly. The entire furnace is enclosed in a metal chamber with silica windows along the light path. The furnace is heated electrically; approximately 4 kW of power (400 A, 10 V) are needed to raise the temperature of the furnace to 2900 K. To aid in vaporizing the sample, an auxiliary arc (70 A, 15 V) was struck to the sample electrode. A collimated beam from one of several primary light sources is passed through the furnace, and the atomic absorption is recorded. The original furnace was lined with tantalum foil to prevent the sample from diffusing through the furnace walls. The later versions were lined with pyrolytic graphite, and the furnace was also filled with argon, usually at atmospheric pressure, to prevent diffusion. The primary light sources consist of hollow cathode discharge 10 tubes, microwave excited electrodeless discharge lamps (EDL's), and a deuterium continuum source. The radiation from any two atomic line sources is combined by a static beamsplitter, and the combined radiation is alternated in time with the radiation from the continuum source. Another beamsplitter is used in conjunction with two mono- chromators so the output of each monochromator alternates between the atomic line and the continuum. Two photomultipliers are used to convert the radiant power of both beams to a proportional electrical current. The photocurrents are sent to a series of sample-and-hold amplifiers which are syn- chronized to the signal. The difference in absorption of the atomic line and the continuum is considered to be directly related to the atom concentration in the furnace, since any non-specific molecular absorption will be subtracted. This process is repeated for the second wavelength, and an absorbance ratio is obtained by a log-ratio type circuit. The second channel and ratio circuit permit the use of an internal standard. The inclusion of an internal standard allows the precision to be improved from 10-25% relative standard deviation to 5-8% relative standard deviation. These data are representative of the precision when a solid is placed in the furnace. If a solution is used, the precision improves by a factor of approximately three. L'vov has prepared calibration plots for 37 different metals. The limits of detection, defined as the amount of sample needed to produce 0.05 absorbance units, ranged from 5 x 10'9 g for boron to 6 x 10"14 g for cadmium (6). L'vov feels that these detection limits 11 would be improved by a modern spectrophotometer which could resolve changes of 0.001 absorbance units. The effects of the matrix on the absorbance were determined by preparing manganese standards in various bases. Even a 104 fold excess of NaCl, Pb(NO3)2 and Sr(NO3)2 produced no significant change in the absorbance of the manganese standard. The calibration plots for most of the elements are markedly non-linear, even over a narrow concentration range. The shape of the working curve is unaffected by concomitants, so one calibration plot can be used for the same element in different matrices. The L'vov furnace is not limited to the determination of heavy metals. L'vov and Khartzyzov (22) used the graphite furnace to determine phosphorous, sulfur, and iodine. They used a vacuum UV monochromator and purged the monochromator and furnace with argon. The excitation sources were microwave excited EDL's made in fused silica tubes. All the other optics were lithium fluoride. L'vov reported sensitivities in terms of absorbance units per gram, so the detection limits have been calculated assuming that 1% absorption (0.0044 absorbance units) is the minimum detectable signal. The detection limits ranged from 4 x 10"0 g for sulfur at 180.71 nm to 9 x 10'12 g for phosphorous at 178.3 nm. Furnace temperatures were 1800 K for iodine and 1900 K for phosphorus and sulfur. The L'vov furnace has many advantages necessary for a nonflame atomizer, but it is very difficult to use. Only a few samples may be analyzed each hour, the power consumption is large (approximately 7.5 kW) and the atomizer is bulky. The importance of the L'vov 12 furnace cannot be overestimated. It is indeed fortunate that the first nonflame atomizer investigated performed so well, because it illustrated many of the fundamental advantages of nonflame atomiza- tion with such remarkable results. L'vov has summarized the analyt- ical results and investigations into physical phenomena with the graphite furnace (23). The applications in fundamental research include the measurement of the Lorentz widths of resonance lines, determination of the absolute values of oscillator strengths, and the determination of atomic diffusion coefficients. 2. The Massman Furnace The King furnace was modified in a slightly different manner by Massman (24) for use in AA and AF spectrometry. Massman's graphite cuvette is similar to L'vov's furnace, but the analytical sample is placed directly into the tube, rather than into a separate electrode. The cuvette intended for use with AA spectrometry is simply a graphite tube 55 mm long, 6.5 mm i.d., and with walls 1.5 mm thick. The fluorescence cuvette has a slit cut in the side for observation of AF. The graphite cuvette is heated to 2900 K within a few seconds by a 400 A power supply. The cuvette is either purged with argon between samples or enclosed in an argon filled chamber, if atomic fluorescence is to be measured. Massman used modulated hollow cathode discharge tubes for AAS in order to discriminate against the background. To obtain an AF signal, the hollow cathodes were pulsed with high current, low duty-cycle pulses. The Massman furnace accepts 5 to 200 p1 of liquid 13 (5 to 50 ul in the AF cuvette) and up to a 1 mg solid sample. Larger samples gave unacceptable background absorption effects. Even the background correction system (25), which determined the atomic absorp- tion signal by computing the difference in absorption of the resonance line and a nearby fill gas or nonresonance line, could not correct for very large background signals. The limits of detection ranged from 4 x 10"4 g for zinc (by AFS) to 2 x 10’9 g for selenium (by AAS). Relative standard devia- tions were from 4 to 12% and matrix dependent. The precision was improved by using a two-channel spectrometer and an internal standard. 3. The Woodriff Furnace The L'vov furnace was modified by Woodriff (26) who made it larger (150 mm long) and used a constant flow of argon through the furnace. The Woodriff furnace requires sample introduction through a sidearm, but can be modified to accept a nebulized sample carried by the purge gas. The tube furnace is electrically heated to a maximum temperature of 3300 K. The sample in the sidearm is atom- ized and then swept into the heated furnace by the argon purge gas. The analytical data show detection limits from 10"] g to 10'10 g for 15 elements. The calibration plots are linear over at least one order of magnitude; for example, absorbance is directly propor- tional to the absolute mass of lead from 0.1 to 20. The precision was not reported, but Woodriff recommends at least triplicate analyses. The optics and electronics have been modified to allow back- ground correction by a unique method (27). Collimated radiation 14 from a line source and from a continuum source are combined in a Glan-Turner prism, which also polarizes each beam. The combined beams are then passed through the furnace. At the exit of the monochromator are two polarizers which move in and out of the beam. The continuum absorption is sent to a reference channel and sub- tracted from the atomic absorption. The precision for the analyses obtained with background correction was not reported but the cali- bration plots in the later papers are curved. When Woodriff used the sample electrode to collect airborn particulates, by forcing air through the electrode, the calibration plot for lead was totally curved, from 0.5 to 8 ng, but was still usable in that region (28). 4. Other Furnaces Numerous furnace methods have been proposed for atomization, but have not been as completely characterized as the three furnaces discussed previously. These other furnaces will be divided into two groups on the basis of construction. One large group of furnaces is heated directly. The other group of furnaces are constructed from a dielectric material, such as silica or alumina, and wrapped in a heating element. a. Furnaces Heated Directly. This category includes the only widely available commercial furnace designed as a nonflame atomizer (29). The furnace, distributed by the Perkin-Elmer company, is called the Heated Graphite Atomizer (HGA) by its manufacturer. It is very similar in design to the Massman furnace, but only atomic absorption can be measured in the HGA. The commercial version has 15 a channel cut into the furnace so that a sample of 5 to 100 pl may be reproducibly placed in the furnace. The furnace can also atomize finely divided powders. The HGA is heated by a two-step temperature program. First, current through the furnace is adjusted to desolvate and to ash the sample. Finally, the current is readjusted to atomize the sample. The furnace is heated to 2900 K by dissipating about 5 kW (400 A, 12 V) in the furnace. Not only is the HGA similar in design to the Massman furnace, but the detection limits and precision are also similar. A matrix study on atomic absorption measurements with the HGA has been completed (30) and several severe interelement effects have been found. Zinc was used as a test element and "all the ions tested caused interference" (30). Zinc is a very volatile element and prone to interferences, however. The reproducibility for 5 ng of zinc was 6%. The authors concluded that a temperature gradient was responsible in part for the poor precision. The HGA was applied to the determination of trace transition metals in seawater (31), and an extraction procedure was found to be necessary for some samples. The reproducibility for samples in a seawater matrix was also 6%. The Massman furnace is the basis for another commercial furnace, but in a miniaturized form. This furnace, the Varion Carbon Rod Atomizer, is also restricted only to AA measurement. A graphite tube about 10 mm long and 3 mm i.d. is held between two graphite support rods. The sample, 1 to 5 ul is inserted into the CRA with 16 a teflon-tipped microsyringe. This teflon prevents the syringe from scarring the atomizer, so precision is increased. The reported rela- tive standard deviation is 1.6% for AA measurements on a 50 pg sample of cadmium in water (32). Other matrices, including leaves, bovine liver, caviar, and pears were examined. The samples were merely diluted in nitric acid, and the results agreed with independent analyses. The precision was good; the relative standard deviation was on the order of 5% for the various matrices. A very simple graphite tube atomizer has been used for the determination of lead in blood (33). The design is similar to that of Massman (24), but the system is simplified by eliminating any sheathing or enclosure. A sample is placed in a tube, and dried on a hot-plate. The tube is then clamped in the atomizer and heated electrically, in two steps. The first step ashes the sample and the second step atomizes it. The relative standard deviation of AA results was 10.5%. The graphite tube was discarded and another one snapped into the holder for the next analysis. Both the sample and the atomizer required some pretreatment. The graphite tubes were degassed in an argon atmosphere prior to analysis. After a 20 ul sample of whole blood was added and dried, 80 pl of 30% hydrogen peroxide was added. The oxidized sample was then dried on a hot plate two minutes, and the tube was ready for the atomizer. The results of the AA technique agreed quite well with the results from a complicated solvent extraction procedure. The 10% relative standard deviation and the cost of replacing the graphite 17 tubes prior to every sample were considered minor disadvantages when compared to the sensitivity and ease of determining lead samples in a complex matrix. The graphite furnaces described are in general large and bulky, with the Varian CRA and its small graphite tube being the exception. The atomizers all require water cooling of the electrical contacts, and require about 6 kW of power to be heated to 2900 K. A different approach was chosen by Robinson (34) who coupled a 6 kW radio-frequency generator to a bed of carbon particles by an induction winding. The analytical results were affected by the matrix and exhibited poor precision, probably because the maximum temperature was only 1700 K. The RF furnace does, however, allow the analysis of gases for cadmium, zinc, and mercury (35). Since Robinson's furnace simultaneously desolvates, ashes, and atomizes, a background signal due to nonspecific absorption may exist. Robinson corrects for this effect by looking up the background signals in a table (36) and subtracting the value from his results. A similar furnace has been used by others (37),but modified to give a temperature of 2700 K. Even at this temperature,matrix effects were prevalent, but the analyses could be easily performed on solid samples. The standards were made in the same matrix as the unknown, and even though calibration plots were curved, they were still usable. The relative standard deviation of 10 AA analyses was 7%. b. Furnaces Heated Indirectly. Although electrically heated graphite furnaces probably have the widest utility, furnaces made 18 from other materials have been used in specific instances. Fuwa and Vallee (38) used a Vycor tube wrapped in heating tape as an atomic vapor reservoir. The actual atomization was accomplished by aspirat- ing the sample into a chemical combustion flame. The atomic vapor and flame gases were directed into the furnace, and atomic absorp- tion measurements were made in the furnace. Absorption in Fuwa tubes was shown to follow Beer's law as the furnace length was varied from 1 cm to 70 cm. The detection limits for the six test elements, including both volatile and non- volatile elements, were all better than found in flame AAS. Matrix effects are present in Fuwa tubes (39), but are different than in flames. The formation of refractory oxides, for example, does not occur in Fuwa tubes, but the presence of volatile inorganic salts produces a background absorption due to absorption by molecular vapor. The presence of molecular absorption prompted Koirtyohann and Pickett to propose a method of background correction which has been widely adapted (40). Their method has been previously mentioned, but not in detail. They essentially performed two absorption mea- surements. In the first measurement, a narrow atomic line source is used; the total absorption measured is due to the sum of the atomic absorption and molecular absorption over the width of the line source. The second measurement is made with a continuum source. The source line width is now much wider than the absorption line, and even if the atom concentration is high enough to completely absorb the radiation over the absorption line width, it is only absorbing radia- tion over a width of about 0.001 nm. Thus, the atomic absorption 19 can be considered negligible, and any absorption measured with the continuum source may be considered to be molecular absorption. The original work of Koirtyohann and Pickett involved two separate measurements, but the method is easily modifiedikn~use in conjunction with an AC system. A chopper can alternate between the line source and a continuum source. The AC signal is sent to a difference amplifier and demodulator so that the net atomic absorp- tion may be recorded in "real" time. Fuwa tubes are still being used. They were recently applied to the determination of trace metals in silicate rocks (41). Since the sample matrix had to be chemically treated prior to analysis, a solvent extraction step was added to eliminate the bulk of the matrix. The added sensitivity of the Fuwa tube avoided extensive extraction procedures, which would have been necessary even if a long path length flame were used. The experimental relative standard deviation of AA measurements was 1 to 5%. Analysis of known standards showed that the errors were random so that the accuracy and precision could be considered identical. Another approach to nonflame atomization is to use a flame to heat a tube furnace. This approach was taken by Delves (42) who used this technique to determine lead in whole blood. A 10 pl sample of blood was placed in a small nickel crucible, oxidized with hydrogen peroxide, and dried on a hotplate. A nickel tube was placed on an air-acetylene flame and the crucible was placed in an opening at the bottom of the furnace. The sample was atomized by the thermal energy of the flame which has been transferred 20 to the nickel tube and crucible. The AAS results were comparable to those obtained by a spectrophotometric analysis,which required 0.5 ml of blood. No background correction was found to be necessary for the AA spectrometric method. The calibration plots were linear from 10"9 g to 2 x 10'8 g of lead which corresponds to concentra- tions of 10 09/100 ml to 200 ug/lOO ml of whole blood. The upper end of the "normal" range is 36 pg/lOO ml (43) and the "dangerous" level is 80 ug/lOO ml (44), so the sensitivity of the Delve's cup technique is sufficient for clinically significant levels of lead. 8. Filament Atomizers Atomizers which have no provision to confine the atomic vapor within the atomizers have several advantages when compared to furnaces. They are usually less bulky and require less power to achieve the same temperatures as furnaces. The major disadvantage of filament atomizers is that the atomic vapor has a chance to cool,and con- densation effects may be seen if other elements are present. A large fraction of the research performed with filament atomizers has been concerned with methods to reduce the interelement effects. 1. The West Atomizer T. S. West and co-workers have authored a series of papers on a carbon filament atom reservoir (CFAR) (45-57). The atomizer started out as a carbon rod 40 mm long and 1-2 mm in diameter. The CFAR was enclosed in a Pyrex housing with silica windows. The fila- ment-type construction allows observation of AF as well as AA, so 21 enclosures were also made with silica windows placed to allow the observation of fluorescence at right angles to the excitation beam. The chamber was filled with argon flowing at 3.8 t/min. The sample, approximately 5 pl, was placed on the atomizer and 500 W (100 A, 5 V) was dissipated across the CFAR. The sample was atomized in about 5 sec, and the atomic vapor produced an absorption or fluores- cence peak. Even though the electrical contacts to the carbon rod were water cooled, the atomizer took two minutes to cool sufficiently to place the next sample on the atomizer. The peak absorbance or fluorescence was plotted as a function of concentration for two elements, silver and magnesium. The cali- bration plots were linear over about one order of magnitude, and the -10 detection limits of both elements were 10 9 when AA measurements were used. Detection limits improved to 3 x 10'11 16 g for silver and 10- g for magnesium by AFS. Hollow cathodes were used as excitation sources for AAS and high intensity hollow cathodes for AFS. The relative standard deviations were usually about 9% increasing to about 30% near the limit of detection. The AA and AF peaks were recorded on a slow (1 second full scale response) recorder, and the electronic distortion was thought to cause curvature in the calibration plots. The recorder was re- placed by an oscillograph and a laminar sheath of argon was used to shield the CFAR (46). A small notch was cut into the filament to ensure reproducibility in sample placements. To perform an analysis, the sample was placed on the filament with a glass micro-pipette and the atomizer was "programmed" to first desolvate the sample, 22 then atomize it. The current to the atomizerwas switched on and off for precisely 90 seconds at the fastest rate achievable by the operator. The atomizer temperature rose to approximately 150° C during this step. The full power was then applied for 2-3 sec. This heats the atomizer to 2900 K with 700 W (10 V, 70 A) of power. The calibration plots were linear over 2-3 orders of magnitude -11 -14 and detection limits ranged from 10 to 10 g for magnesium, silver, lead, zinc, bismuth, thallium, and gallium. These elements were all determined by AFS with high intensity hollow cathode lamps as sources. The detection limit for magnesium was 10'12 than the previously reported 10'16 g. The relative standard devia- 9 rather tions were less than 2%,which is surprising since the precision of the sample delivery was estimated to be 2%. The apparatus was further modified (47) by removing the enclosure. A detection limit of 1.5 x 10"3 g (or 0.15 ng/ml) was observed for cadmium by AFS after a light guide was installed to prevent the back- ground emission from the CFAR from reaching the detector. Unfor- tunately, almost every conceivable interference was present as an interelement effect. The determination of gold (48) was the first time that West's CFAR was used to determine an element not considered to be volatile. Gold has a melting point of 1336 K and a boiling point of 2873 K. The samples were atomized sufficiently well that a detection limits\t)]d>\ (2) where 2 is the path length for absorption and RA is the atomic absorption coefficient. The atomic absorption coefficient can be described in terms of fundamental parameters if the line width can be assumed to be governed by the Voigt profile expression. This is generally the case for a single isolated spectral line 38 mOhflzomoOZOE .Frmo Loam> ow20p< 85F ._ mesmvm momDOm 39 where broadening is primarily Gaussian (Doppler broadening) and Lorentzian (collisional and natural broadening), and kA can be written aK "fOu exp(-y2)dy kx = 7‘91?“ 7—2 (3) .00 a +(U‘y) where 2/2n2 (A-A ) v = AA 0 (4) D /£nZAAL a = AAD (5) 2A _ o [2(tn2)kT AAD - e V M (6) Z/JthXAOZ K0 = —————-— (7) ffemD x = 123 (8) me where A is any wavelength in cm; A is the wavelength at the line 0 center, cm; M is the atomic weight of the atom, g; fOu is the absorption oscillation strength of the ground state (0) to upper state (u) transition, dimensionless; m is the mass in grams and e is the charge, in esu, of the electron; c is the speed of light, 1 cm sec' ; AID is the Doppler half-width; AIL is the Lorentzian 4O half-width; a is the classical damping constant, dimensionless; v is a variable dimensionless wavelength interval taken with respect to AID; y is a dimensionless integration variable; n is the atomic con- centration in atoms cm'3; and k is Boltzmann's conscant. Note that the atomic absorption coefficient kA includes n, the atomic concentration. Equations (1) and (2) show the dependence of the absorbed radiance on the integrated absorption, A The dependence is best t. seen in two limiting cases: A = [l—exp(-kxt)]dx 3 Blikxdx (9) When the optical density, i.e. n2, is small. When the integration is carried out, : V0 KOfRAAD A - n (10) t 2711? at high concentrations, however, fiKftan At = V—m—Q’ “’7 (‘1) So at low concentrations, A is proportional to n, the atomic con- i: centration. At high concentrations, however, A is proportional to t the square root of n. At intermediate concentrations, A is dependent t on a, the damping parameter. 2. Atomic Absorption with a Line Source If a line source is used for AAS, the radiance absorbed, BAAL is given by 41 BAAL = BL[l-exp(-kmt)] (12) where km is the average atomic absorption coefficient at the absorp- tion linecenter.llfis implies that the line width of the source is less than the absorption half-width. The opposite assumption is implicit in the derivation of the equations governing the continuum case. BL is the radiance of the line source in W cm'2 sr‘]. For low concentrations of the analyte (i.e. small km ) B E kmts (13) AAL L since km is related to the atom concentration, n, the absorbed radiance is directly proportional to atom concentration at low concentrations. At high concentrations, however BAAL E BL (14) The physical significance of this equation is that at sufficiently high concentrations, all the excitation source radiation is absorbed, and the absorbed radiance is independent of'concentration. 3. Radiance Expressions for Atomic Fluorescence Fluorescence is considered to be isotropic, thus occurring in all directions with equal magnitude. The radiance of atomic fluorescence is obviously proportional to the fraction of primary source radiation collected by the atomic vapor cell, which is O/4n where O is the solid angle collected and 4n is the number of sr in a sphere. BAF is proportional to the radiance absorbed, BAA’ and also proportional to the number of atoms in the plane L x t' of 42 Figure 1. This plane is directly in line with the monochromator. The observed fluorescence is inversely proportional to the cell depth and height because so many of the planes will be out of the field of observation of the monochromator. The observed fluores- cence is finally proportional to the quantum yield, Y, as shown in Equation 15. th' =.£LB 4n BAA txt' BAF Y (15) There is, however, a chance that some of the fluorescent radiation will be reabsorbed by the atoms between point of fluorescence and the monochromator. This term, the self absorption factor, F, is given by the ratio of the radiance absorbed in the emission path to the radiance in the emission path as shown in Equation 16. _ At(nL) —;;————- (16) f0 kALdA At low optical densities (low nL), F can be reduced to f:[1-exp(-kAL)]dA~ 7:“de f0 2,de f0 kALdA For high optical densities, F is given by [ ( )] ‘J Jrk anaADZ ;f 1- k AL )dA exP 102 _ .._l_ (18) f0 kALdA v/l-T—K onfLAAD \/ Eli/503513 fri 2/tn2 n Thus the self absorption factor will increase with the square root of concentration under high concentration conditions. The final expression for the fluorescence radiance is Q L %F meiYF “” When log BAF is plotted as a function of log n, as in Figure 2, several points become obvious. At low concentrations, the rela- tionship is linear, with unity slope. At high concentrations, the slope levels off'to zero,and fluorescence becomes independent of concentration for a continuum source. This happens because the 1/2 absorbed radiance becomes proportional to n at high concentra- tions,while the self-absorption factor, F, becomes proportional to n-1/2. When a line source is employed, the fluorescence actually decreases with increasing concentration at high concentrations. The physical significance of this is that all the source radiation is absorbed, and the excited state atom population has reached a maximum. As concentrations increase beyond this point the self absorption becomes dominant and the slope of the log-log plot ap— proaches -l/2, which clearly shows the significance of the 0’”2 term in the self absorption expression. Figure 2 has been redrawn from Winefordner, Svoboda, and Cline (84); this excellent review is also the source of the radiance expressions presented here. INTENSITY RELATIVE FLUORESENCE 44 10"1 CONTINUUM (/~€ 0.01 1 I w 1 no .60 1.000 , 10.000 ATONIC CONCENTRATION Figure 2. Growth Curves For Atomic Fluorescence (84). 45 C. Characterization of the Atom Population of a Nonflame Atomizer The preceding expressions examine the dependence of RAF on n, the atom concentration. When using nonflame atomizers, the atom population in any volume is a function of time. The following derivation is from L'vov's work (86). 1. Time Dependence of Atom Population If a(t) is the rate of atomization and c(t) is the rate at which atoms leave the cell, then the atom population may be des- cribed by = a(t) - c(t) (20) 0.0. (*2 where N is the atom concentration at any time. L'vov chose to assume that the atomization rate is directly proportional to time, a(t) = At (21) The coefficient A may be evaluated by integrating the rate of atom- ization over the time of atomization, TA 1' .[ Aa(t) = No (22) 01" 2N a(t) = —§ t (23) TA where No is the total number of analyte atoms in the sample. The rate at which the atoms leave the cell is 46 _N_ Tc c(t) = (24) where Tc is the mean residence time of an atom in the cell. If Equations (22) and (23) are substituted into equation (21) the result is R=—7‘? W” A pair of equations results if the variables are separated and the Equation (25) is integrated as done in Reference (85). 2 2NoTc t N = TA L;— -l+exp(-t/TC)] for t.: TA c 2N0Tc2 A N = -j;7§——{;;--l+exp(-TA/tc)]exp[-(t-IA)/tc] for t 3_TA (26) A . The dependence of N/No, the relative atom concentration, on time is shown in Figure 3. The time axis is in units of TA, the atomization time, so the plot demonstrates the effects of restrain— ing the atoms in the observation cell. The results are clear: if the mean residence time spent in the observation can be increased, the signal will increase. The residence time in the observation cell, however, is gen- erally related to the design of the atomizer and invariant from sample to sample. The atomization time can be varied, usually by varying the temperature of the atomizer. When the residence time is assumed to be constant, and the atomization time is varied, the resulting dependence of atom population on time is plotted in 47 LO- 2 9 '2 40.8r- :3 o. o a. goo L l.— < ‘9' "O.4r E .J 1.1.! a: 0.2r 2' 4 6 8 IO TIME, MULTIPLES or ATOMIZATION TIME Figure 3. Dependence of Atom Concentration on Residence Time in Observation Cell. 48 Figure 4. For very short atomization times, the peak population approaches the sample population,but the peak signal is very de- pendent on the atomization time. For example, if TA/TC = 0.1 then N = 0.95, but if TA/Tc = 0.2, then Npeak decreases to 0.90. peak 2. Time Integrated Atom Population Expressions If the integrated population, 0, is measured, then TA N T m N T Q =1 3 CU-exm-t/rcndt +f ° Cn-expoAn )1- o A TA TA C 8Xp[-(t-TA/Tc]dt (27) 01" Q = N I (28) This shows that the integral method provides an exact measurement which is independent of atomization time. The sensitivity of the integral method may be increased by increasing the residence time of the atoms in the observation cell. If the residence time is constant, as in the large majority of nonflame atomizers, then the integral method becomes independent of the atomization param- eters. I The peak method depends on the atomization time being constant. This may not be a good assumption for samples containing a large concentration of the analyte, or even a small concentration of the analyte and a large concentration of other elements. 49 1.0 .. Zoe .. 9 .— 3 a A rC/TA = 5 00.6 )- a. g B TC/TA = 1 '2 0.4 ) LIJ _>.. '— 3 3:02 A c TC/TA = 0.2 2 4 6 8 1‘0 TIME,MULTIPLES OF RESIDENCE TIME Figure 4. Dependence of Atom Concentration on Atomization Time. 50 0. Time Dependence of Temperature The atomization of a sample occurs as the atomizer temperature is changing from a low initial temperature to a higher, steady state temperature. L'vov assumes that the temperature (atomization rate) is linearly related to time over the atomization region. This assump- tion will be examined in detail in later sections. 1. Heat Sources and Heat Sinks in a Filament The fundamental hypothesis describing conduction of heat in isothermal solid is 2 fl:K§_; (29) 3x 3 where T is temperature; t is time, K is the thermal conductivity of the solid, and x is the distance along any arbitrary axis. This equation will not describe the temperature of an electrically heated filament as normally used for nonflame atomization. Effects from the edges will be ignored. The atomizer is heated by electrical current at a rate R given by . 2 R = 417 (30) pcA G where I is the current; j is a conversion from joules to calories; 0 is the density of the filament, A is the cross-sectional area of the filament; c is the heat capacity; and G is the electrical conductance. Another source of heat arises from the Thomson effect which can be expressed as 51 = 15 EU: R 'EEA'BX (3]) where s is the Thomson coefficient for the filament. Heat losses are due to conductive losses or radiative losses, if convective losses may be ignored. The rate of heat loss due to conduction can be expressed as 'R= pHCA (T T 0) (32) where H is the surface conductance; p is the perimeter; and T0 is the temperature of the medium surrounding the filament. Radiative heat losses are given by = 111—'59 (14-104) (33) where H' is the product of the Stefan-Boltzman constant and a func- tion related to the emissivity of the atomizer and the surroundings, and K is the thermal conductivity of the atomizer. When all the terms are summed, Equation (34) results . 2 . (34) + I %UF’(T4’T04) pcAzG where K/pc =1<. The electrical conductance and thermal conduc- tivity are both functions of temperature, as shown in Equation (35) and (36) and must be included in Equation (34). % E]; (1+dT) (35) 7Q II Ko(1+eT) (36) 52 where Go and K0 are the initial electrical conductivity and thermal conductivity, and a and B are arbitrary parameters which provide the best fit to the experimental data. When Equations (35) and (36) are substituted into Equation (34), the result is 3T- = £9 i[(1+BT)—°—T- -S——I 331-11 T + ———HpT° 3 pc 3x 8x Apc 3x pCA pcA . 2 . (37) -J%%—(ndn ”HmTA(14-104) pcA Go If the only information desired is the variation of temperature with time at points well away from the edge,then %%-may be set to zero because the filament will be of uniform temperature. This leads to t ocAZGo pCA 1+BT (38) HpT . 2 1 A) *"Tu2“ DC pCA G If the terms with similar dependencies on T are gathered together, Equation (39) results. 4 )+ C' (39) 4 T -To dt 3 (1+BTO 2. Zero Order Perturbation Terms Equation (39) describes the variation of temperature with time at the nonflame atomizer, without convection terms. It cannot be solved exactly, but an approximation might be sufficient to des- cribe the relationship of time and temperature. The perturbation 53 method of Horne and Anderson (87) was chosen as being the most promising method of simplifying the equation. In this method, estimates are made for the coefficients of the various terms, but small terms are not ignored; rather they are included as perturba- tion terms so that the effect of possible errors in the initial estimates are minimized. Equation (38) can be rewritten as 4 4 91.: [0A -A ]T + B :—::9— + A +A T (40) dt 1 2 1+8T l 2 o where . 312 pcAzGo 2 - pCA If T :1000 K at the time of atomization then A] z 108B and A2 2 1058. To aid in scaling the coefficients and making rational judgment concerning their magnitudes, let T = T/lOOO and B = B'/1000. Then Equation (40) can be written as d /1000 1012T4_m12T 4 —ljfif——-= (aAl-A2)(1000T) + B 1+B.T ° + A1+1000A2To (41) or Q. T 4 4 T‘T "E = (0A1-A2)T + AZTO + A] + 1012 am???) (42) 54 If 10123 is factored out of Equation (42), Equation (43) results 4 4 A T 'T oA -A 2 A gi" 10123 (Tis'o) + 1122 T + 12 To + i2 ! (43) T 10 B 10 B 10 BI Approximate ratios of the four terms in brackets are 1 : 10'7 : 10'4 ; 10‘7 Now a perturbation expansion (87) may be introduced since the ratios are known. The four terms in Equation (43) will be multiplied by e to the appropriate power in order to keep track of their relative magnitudes 4 4 T -T (GA -A ) A A -—= .0 ° +3 -———-—— +e———i2 .0 + ,__2 1. 1 44> 1+B'T 10 B 10 B 10 B The temperature term is expanded in a Taylor series. 2 = €°T¢ + ST] + €2T22 + €3T33 + .... (45) The parameter c is simply a bookkeeping device which aids sub— sequent manipulation of Equation (45). When c is 1, then i = T in Equation (45). After the appropriate mathematical operations are completed, 6 will be set to unity; in the interim, however, the power of c is used to aid in making rational decisions about the relative sizes of the various coefficients. As in any perturbation scheme, one cannot be certain a priori whether the solution of Equation (45) converges when 8 = l. Hopefully Equation (45) does converge rapidly and may be truncated with minimal error. The T¢ term repre- sents the zero order perturbation and is not to be confused with To, 55 the initial value of T. The first step is to collect the zero order perturbation terms dT T T __2.= 10123.J2__Jl_ (46) dt 1+B'T d where T4 has been evaluated from Equation (45) and all terms of order 5, 82, etc, have been ignored. Equation (46) may be solved exactly: 4/21 3 T 2+1 T J2+T 2 T T /2 - 0 d 0,9 0 o d t -——————- log + 2 arc tan ———————— 1012B T 2-T T J2+T T 2-1 2 p o d o o p 2 (47) 14 + (8'2/210)arc tan -——§ To To extract T¢ from this equation, nocollect the first order per- turbation terms, and then solve exactly the first order perturba- tion equation is impossible. Equation (46) may be rewritten with the denominator expanded in a Taylor series, and truncated after one term 1+8'T¢ 12 -————-—-dt = 10 Bdt (48) T 4+T 4 o p 0 or 12 “3'1 10 Bdt - T 4(]-T013 dt¢ (49) 0 "‘4 T ¢ Now, let T x = O §_x §_l (x 81/3 to 1/4) (50) 56 and substitute into Equation (49). 1+B”-Q (-Io)dx 10128dt = 4. x 2 (51) To 4 X --(1-x ) x4 (x+B'T )x 10128dt - 3 04 dx (52) T (1—x ) O x(x+8'T ) 1012Bdt = - 3 ° (1+x4+x8+...) (53) T O The series arises from the expansion of (l-x4)']. Since x<%, then x4 and higher order terms may be truncated and still be accurate within a few percent. Equation (53) is easily integrated to produce T 2 -101ZBTO3t = %x3 + %B'To—2§'+ constant (54) "‘15 If Equation (50) is substituted into Equation (54), -10]28t = 1 3 + B; + constant (55) 3T¢ 2T¢ The boundary conditions demand T¢ = T0 at time t = 0, so the constant may be evaluated as ‘3 + B; = -10‘28t + I 3 + 2 (56) 3T¢ 2T¢ 3T0, 210 This equation can be "simplified" to T 3 _ 8'T T + l | = O (57) 310 210 310 2T0 57 Equation (57) has one real and two imaginary roots. The real root is = ,1. to); "3 I_1_ [1,.)(51311/3 To 6M 2 1 6M ‘V36M 3 (58) 36M 216M3 216M I where 6x1012BTo3t-2+3B'T0 M = 3 (59) 6T0 If temperature is related to time, then 6x10128Tot>>1 (60) or else M becomes almost independent of t. Note that this implies 12 that B>6x10 is true. This cannot, at the moment, be documented. Under these conditions, [i.e. for (60) true], T¢ can be written 1 1/3 T 1 ° 3.6x10‘38t (61) The zero order effect is the radiative loss of heat. It is obvious that this cannot be used as an approximation since conductive losses and Ohmic heating do not appear until the first order terms are included. 3. First Order Perturbation Terms Since I” is known, first order perturbation terms may be collected. 4 4 3 E:l _ 12B (To 'To ) . 4T2 dt ‘ 1 ‘0 2 B ' (62) U(1+8 T”) (1+8 T¢) 58 The solution to this differential equation is I] = exp(K]t5/3 + K2t1/3) (63) where K = 9x106B2/3+6.6 1 8' 1.44x10'2t048'”3 K2=- B. The assumption 1+8'::1 was made to solve Equation (62). This con- tributes no more than 5% error over 300 to 1100 K. Now if the zero and first order perturbation terms, are summed T = T¢ + T] _ _ 6 2/3 T = E 1/3 t 1/3 + exp [9x10 3' +6.6 t5/3 3.3x10 B 1.44x10‘2104+ 8"” 1/3 T = c1t'1/31-exp(c2t5/3 + c3t1/3) (65) It is known that during the first part of the atomization step, 5/3 term must be dominant; the temperature is increasing, so that the t the coefficients of the other terms are negative and represent tem- perature losses. If Equation (65) is expanded and truncated after one term. T 6 t5/3 (66) 59 The nature of the dependence of temperature on time seems 5/3. This term strange; few natural phenomena are related by t is actually a combination of two terms. If only heating effects are considered .9: = const dt l+oT for Ohmic heating. This is easily solved. 1 = At + 8t2 Note that Equation (66) also shows that the dependence of T on time 5/3 is between a linear and a squared dependence. The t relation- ship results because the perturbation approach gave relative weights 2 5/3 to the t and t terms, and provided a weighted average as t . E. Modified Estimates of Atom Population The differential equation describing the atomic concentration in the vapor cell Equation (20) is still correct 9-2- = a(t)-c(t) (20) where a(t) is the atomization rate and c(t) is the rate at which atoms leave the cell. But a(t) was shown in Equation (66) to be proportional to ts/3 5/3 a(t) = At (67) The coefficient comes from the normalization equation TA .1[ a(t) = NO (22) 60 and can be shown to be N 3 o A = -—-—- (68) 8 A Equation (20) can now be rewritten as N dN _ 3 0 5/3 N 7‘s?“ ’1— A c A solution to this equation is N=-—N2—t8/3ex(-t/ ) Oo>_4 Lo mcowpo_umna _au_umcomce mo cemwnmasou .o mtzmwa mic. ZO_Po>_4 w. s. «a CD (3 CD NOLLV'IndOd WOlV HALLV'IBH 0°. (3 O.— 63 shape of the predicted relationship quite well. The addition of the factors contributing to heat loss would tend to flatten the curve at the top, and the fit would be better yet. IV. INSTRUMENTATION Many of the problems uncovered by researchers using nonflame atomizers are caused by improper choice of atomizers and measure- ment instrumentation. If the process of atomization is to be studied, some effort must be made to choose the atomizer most conducive to the study, and to utilize the appropriate measure- ment technique. Atomizers designed specifically for analytical applications, for example, may not be suitable for use in a funda- mental study of the atomization process. When this study was started, the lack of any commercial system necessitated the con- struction of a nonflame spectrometer. The platinum loop atomizer of Bratzell, Dagnall, and Winefordner (77,78) was the first choice for an atomizer. The practical factors, such as ease of use, dominated the temperature limitations in the choice. The platinum loop atomizer is not as good for analytical purposes as some other atomizers, but the disadvantages are rela- tively minor. The criteria used to design the nonflame spectrometer are divided into two parts, with the atomizer and the spectrometer considered spearately. The fundamental and practical limitations are presented and an attempt has been made to outline the necessary compromises made in the construction of the nonflame spectrometer. 64 65 A. Design Criteria 1. The Atomizer Only two rigid requirements were imposed on the atomizer. The first was that it be capable of atomizing an analytical sample. The second requirement was that the atomizer must be adaptable enough to be useful throughout a complete characterization. Since the initial results would direct future research, a flexible atomizer was necessary. This eliminated furnaces; their geometry precludes the observation of atomic fluorescence. a. Practical considerations. One important detail often overlooked in the design of the atomizer is the ease of placing the analytical sample on the atomizer. The effects due to sample intro- duction must be isolated from atomization. For maximum sampling precision not only must identical volumes of sample be placed on the atomizer, but the sample must be placed in the same spot every time. Again, this tends to rule out furnaces. Another very important detail in the construction of an atomizer is the ease of temperature control. The best control would obviously be achieved if electrical heating were used. Unfortunately, no resistive element can convert electrical power to heat with 100% efficiency. There are also radiative losses, both in the visible and radio frequency regions of the electromagnetic spectrum. Visible radiation can be tolerated, but electromagnetic interferences (EMI) can cause severe problems. Even a small amount (10'3 W) of EMI may cause any nearby electronic instrument to produce erroneous results. This is an intolerable condition in a research laboratory 66 which contains complex electronic instruments such as digital volt- meters, strip chart recorders, and digital computers located in the vicinity of and sometimes used in conjunction with the atomizer. The radiated portion of the EMI occurs chiefly when an alter- nating current crosses zero. If a 6 kW transformer were used to heat a carbon filament atomizer, like West's (44), the EMI would be quite large. This problem can be minimized by using direct current, but a 1000 A power supply is prohibitively expensive. An alternative solution is to avoid using a large power supply. If the power could be kept below a hundred watts, then EMI would probably be negligible. The carbon filament atomizer was reluc- tantly eliminated due to its high power consumption. A platinum loop atomizer, however, could easily be used under low power con- ditions. b. Fundamental considerations. Platinum metal is considered to be inert to all mineral acids, except aqua regia. Under oxidiz- ing conditions, fused alkali metals will attack platinum but only to a small extent. Molten halides, carbonates, and sulfates have little effect on platinum. The dissolution and separation of platinum is, in fact, a difficult analytical problem. The best methods for dissolving platinum are fusions (88). Platinum is soluble in a zinc metal fusion at 800° C, but the zinc is volatilized at 1000° C. A lead fusion is also used, but takes two hours at 1200° C to dis- solve platinum. These chemical reactions are insignificant to the application of platinum as a nonflame atomizer. Platinum will form a black oxide, PtO, if heated in the presence of oxygen. The oxide, 67 however, decomposes at 550° C. Platinum is also attacked by chlorine gas at elevated temperatures, but all the platinum chlorides decom- pose at temperatures less than 585° C. The major disadvantage of using platinum is that it has a relatively low melting point, 1769° C. This is approximately 600° C less than can be achieved with graphite filament atomizers. This was not considered a severe disadvantage because an atomization study could be made simply by using a volatile element for the study. The platinum loop atomizer has one fundamental advantage over graphite; there are no problems arising from the sample soaking into the atomizer when platinum is used. This can be important if the temporal and spatial characteristics of the atomization process are to be studied. 2. The Spectrometer If all the advantages of nonflame atomization are to be pre- served, then the measurement system must not degrade the improve- ment gained by atomization. Both the optics and electronics may have to be different than those used for flame atomizers. a. Optical considerations. One of the most obvious advantages of nonflame atomization is the minimal background signal. If the atomic vapor is viewed a few millimeters above the atomizer, the background due to the atomizer is essentially zero. This advantage is much more important in atomic fluorescence spectrometry than in atomic absorption spectrometry. In AAS, the signal of interest is the small attenuation (by absorption) of a relatively high source radiance. In AFS, the signal is a small increase above the 68 background. At the limit of detection, the dominant noise source is source flicker noise in AAS, while in AFS the largest noise sour- ces are shot noise in the signal and flicker noise in the back- ground. If a low background nonflame atomizer is used in conjunc- tion with atomic fluorescence measurements, the system will be limited by fundamental noise. This is a very desirable circumstance as the shot noise limit can never be improved. Since the atomic fluorescence signal is directly proportional to the radiance of the primary excitation source (if a line source is used) it is obviously desirable to use the most intense source possible. The spectrometer should also incorporate wide-angle optics in the monochromator. The observed AF signal is directly proportional to the acceptance angle of the monochromator. The monochromator does not have to be a high resolution instrument because the resonance fluorescence produced from atoms is itself highly monochromatic. Hence the light reaching the monochromator due to atomic fluorescence is already monochromatic if a monochromatic source is employed. In practice, however, light reaching the mono- chromator consists of stray light and non-resonance fluorescence as well as resonance fluorescence. The monochromator should be able to separate the desired fluorescence radiance from the stray light for Optimum instrument performance. The major sources of stray light, include room light, and light from the primary excitation source which has been reflected into the monochromator by the atomizer or other nearby reflective surfaces. b. Electronic considerations. The radiation must be transduced 69 into an electrical signal to permit flexible signal processing. It is easier to filter, shape, or digitize an electrical signal than an optical signal. The ideal transducer will be the one which is most sensitive and least noisy. A multiplier phototube, more com- monly called a photomultiplier tube (PMT) has been found to be the transducer of choice for the measurement of low radiant sources. 8 The background (dark current) is approximately equivalent to 10‘ W of radiant power at the input (photocathode surface) which is equiv- alent to about 100 photons sec.1 at a wavelength of about 220 nm. The photomultiplier transduces the fluorescent radiation into a proportional current. The transfer function is linear over ap- proximately 8 orders of magnitude (89), so the amplitude distortion introduced by the detector is minimal. The frequency response of the PMT is considered to be about 20 MHz (89), high enough to respond to transient AF signal (duration in tens of milliseconds) without distortion. The photocurrent should be converted to a voltage and amplified to prevent electrical noise sources from becoming a major source of error. Voltages of approximately lllare also more easily treated than currents of 10'7 A. The signal, now a voltage, is ultimately transformed into a number, but further electronic modifications may be necessary before the digitization step. Many analog-to- digital converters (ADC's) will not digitize a signal properly if the signal is "noisy", i.e. contains noise spikes on t0p of a smooth signal. An ADC which is designed to reject noise spikes may also reject part of the signal. 70 The electronic signal conditioning must be easily varied if the nonflame spectrometer is to be used for a variety of measurements. The electronics will be needed to reproduce faithfully the time— dependent fluorescence, but should also be able to produce a signal proportional to time-integrated fluorescence. The amplitude of the electrical signal must be large enough so that the signal can be digitized with minimal error. The instrumentation used in the study of nonflame atomization was varied throughout the study. Obviously, if modification of the instrumentation improved the results, then the modification was preserved. Since the components of the measurement system affect the electrical analog of the AF signal to a large extent, they were varied in the course of the experiment. 8. Sequence of Events The "normal" nonflame AF analysis consists of a series of separate events. First, the sample is placed on the atomizer. The atomizer is heated, and the sample is atomized. The resultant atomic vapor absorbs radiation from a primary excitation source and the radiational deactivation (fluorescence) is observed. The time- variant fluorescence is transduced into an electrical signal which is suitably modified so that it can be accurately digitized. The ultimate result of every measurement is a number. The instrumentation involved in each of these steps will be described in the following sections. Many of the events are con- trolled electrically, so some sort of master controller must exist 71 to determine the sequence of events. The controller consists of a clock to produce a series of pulses and some other circuitry to guide the pulses to the proper destination. This is necessary to insure that the appropriate sequence of events is followed in every analysis. The design of the master controller is actually trivial. The procedure followed in the actual research was to evaluate the in- strumentation separately for each step, and then to develop the master control unit. The design of the instrumentation assumed that timing pulses could be provided in the correct sequence. C. Sampling to the Atomizer To characterize adequately nonflame atomization required ap- proximately 20,000 separate samples. Even before the study was started, it was apparent that a large number of samples would be necessary. In every study cited in Chapter II, the analytical sample was placed on the atomizer either with a syringe or pipet. The time-consuming nature of the sampling process probably accounts for the lack of research into the nature of atomization. If such research is to be performed in a reasonable length of time, then the sampling process must be automated. The sampler must deliver precise volumes and place them on the atomizer in a reproducible manner. A glass microsyringe is capable of reproducibility with relative standard deviations of less than 2%, in the hands of a skilled Operator. This precision may be lost if the operator accidentally jars the platinum loop atomizer while placing the sample on the atomizer. 72 The first automated sampling system delivered the sample to the atomizer from a quartz capillary positioned beneath the atomizer. The analytical sample was placed in a reservoir and pumped into the capillary in discrete steps by a peristalic pump. The peristalic pump is shown in Figure 7. A movable "wiper" attached to a motor compresses the tubing leading from the sample to the atomizer against a machined radius and forces a sample through the tubing. The precise sample size is determined by the internal diameter of the tubing, the length over which the wiper compresses the tubing, and the distance between the machined radius and the wiper. The latter is variable and was used to adjust the delivery volume from 2 pl to 40 p1 with 0.8 mm i.d. tygon tubing. The capillary construction and position relative to the loop are also shown in Figure 7. The sample clings to the loop rather than the capillary, probably due to increased surface tension at the loop. The height of the tip of capillary must be the same as the level of the sample in the reservoir, or siphoning effects will be seen. The level of the sample was kept constant by a reservoir similar to the mercury reservoir used with a dropping mercury electrode. When the pump was tested, the relative standard deviation for dispensing five 4 p1 samples was 2-4%. The only disadvantages of this system were adsorption Onto the tygon tubing and a large dead volume. After several hours of use, the test element could become adsorbed onto the walls of the tubing. This produced high values when the blank and very dilute solutions of the analyte were run. The adsorption onto the tygon tubing was no particular problem 73 .Empmzm xcm>APmo oFQEwm .m mczmwa Eo>mwwwm w4¢2 J r 2.2:. 28: 74 during the study of atomization when the analyte concentration was kept constant, and at a level significantly above the detection limit. When calibration plots were run, a syringe was used to deliver the sample. 0. Electrical Control of Atomizer Temperature Platinum is readily heated to incandescence by dissipating a few watts of electrical power across the platinum filament. To heat a 32 guage platinum wire to 1600 K requires less than 10 W of power. The resistance of platinum increases as temperature increases. This is a common phenomena in metallic conductors and actually sim- plifies the design of temperature controllers. The term "tempera- ture control“ or "temperature program" is a little misleading. In the study being described, as well as all those cited in Chapter II, an electrical parameter, such as current or voltage, is being con- trolled; temperature is measured. Temperature, current, and voltage are inter-related. If the current and voltage are known,then a resistance may be calculated from Ohm's law. The dependence of the resistance of platinum on temperature is very well known. If a constant voltage is placed across a platinum filament a large initial current will be present, which decreases to a smaller steady-state value because as the filament is heated the resistance increases. If a sample is placed on the atomizer, the atomizer temperature will remain at the boiling point of the solvent until the sample has been desolvated. The temperature then increases 75 to a steady-state value when the Ohmic heating equals the conductive, radiative, and convective losses. If the steady state temperature is chosen properly, the sample will be atomized at some time follow- ing the desolvation, but prior to the time when the steady state temperature is reached. The choice of atomizer voltage is obviously important. If the applied voltage is too high, the Ohmic heating may be much larger than the heat due to desolvation. Under these circumstances the sample will boil furiously, and the possibility of sample loss due to splattering and sample explosion is large. If the applied volt- age is too low, the sample may not be completely atomized. The voltage control to the platinum loop atomizer should be continuously variable so that the applied voltage may be optimized for each element and matrix. The voltage control must also be free from EMI if any digital electronics are to be used in the vicinity. A simple solution is to use an autotransformer (Variac) to set the applied voltage and to use a switch to turn on the volt- age whenever it is desired. This type of circuit will not, however, minimize EMI. The origins of electromagnetic interferences are not well known. The magnitude of the EMI is known to be proportional to the second derivative of current with respect to time (90A). If a switch is turned ON, the current goes from zero to a finite value in a few microseconds. The second derivative is larger yet. If, however, an alternating current is switched ON precisely as it crosses zero, the magnitude of the EMI will be minimized. The switch must be 76 bounce-free or the advantages gained by switching at the zero cros— sing will be lost. No mechanical switch is absolutely bounce free, so a solid-state switch must be used. The only solid-state device capable of switching alternating current is called a bi-directional triode thyristor, or triac (90). A block diagram of a bounce free power switching circuit is shown in Figure 8. The zero crossing switch senses when the alternating current crosses zero, and pro- duces a pulse which straddles every zero crossing. The pulse is amplified and then directed to the gate of the triac. The triac turns ON and completes the circuit containing the platinum loop atomizer. A fraction of the 6.3 V platinum loop power source is selected by the autotransformer and applied across the loop. The low voltage is advantageous because it eliminates the shock hazard of the circuit. The complete circuit diagram is shown in Figure 9. A logic level signal at the input activates the CA3059 integrated circuit which produces a series of zero crossing pulses. The pulses appear at pin 4 of the IC, and are directed to the base of a transistor. The output of the first transistor (2N3393) is sent to a second transistor (2N3053) and then to the gate of the triac (2N5444). The two transistors comprise the amplifier of Figure 8. The 6.3 V source is transformer T2, and the transformer T3 is the auto- transformer. The entire circuit can be considered as an EMI free power supply. When a signal is present at the input, power is gated across the load, in this case, a platinum loop. The loop is heated 77 P p.3oewu Foeucou cmzoa Peace to genomes xooFm .30.. 552.535 mmzmonmzawe .w wgzmwa 10.526 oz.mmomo OmmN .3 .- ..... OOOOOOOOOOO 0000000000000000000000000000000000000 OOOOOOOOOOOOOOOOOOOOOOOOOOOOOO OOOOOOOOOOOOOOOOOOOOOOO OOOOOOOOOOOOOOOOOOOOOOO l eaaeao ..o<_m._... o» 78 .uwzucwu Pogccou emzoa mece we Ewgmmwo pesocwo .m we:m_a Omm \‘LI _V_ E. “ECO—+41. 20.9 ZO<<< mm “.559 mmmon " Jomhzoo . 95¢ . 8.5. " 5238a... 55:35. . mks. . op mo<5o> . . . . r-" '''''' -'|'l" """" hszDmhmE 44:05 442522: now: am 1.55“ 90 atomization time. A ten second time window was chosen to ensure' that the atomization would not be missed. The results produced by the integrator were good when the samples were large and the AF signal duration was several seconds. At the detection limit, how- ever, the signal was less than O.l sec in duration. Under these conditions the background was integrated over 100 times as long as the signal. This situation can be tolerated, but not when the background drifts. The input to the integrator is actually the sum of three components + iAF iin = iBKGD + i0A where iBKGD is due to stray light and photomultiplier dark current; i0A is the input bias current of the OA; and 1AF is the photocurrent from the atomic fluorescence. The input bias current was found to be the largest source of imprecision. Even though a high quality operational amplifier was chosen, the temperature coefficient of input bias current was approximately lO'9 A/°C. If ambient tempera- ture changed by as little as O.l° C, then the input bias current would change by l0"10 A; when integrated for 10 sec the bias current makes the same contribution to the output as an AF photocurrent of 10'7 A for lOO msec. Since the peak photocurrent at the limit of detection was less than l0"9 A, this system was discarded. b. Variable-time integration. One possible method of mini- mizing the amount of background that is included with the signal would be to open the integrator only when the signal is present. ”runny , [fa—m-yx‘l .. m -: ~ ." .l‘ , ‘ i. 91 This obviously requires that the instrument recognize the presence of the AF peak. Peak recognition may be accomplished by a device called a comparator. If the input to the comparator is above a certain level, the output is high. The output of the comparator is used to gate the photocurrent to an integrator so that the integrator is ON only E when the current is above a certain level. This situation occurs {.h.- for two cases; for the AF peak and for noise spikes. Since noise ¥ ‘- spikes have relatively short time duration when compared to the E3 AF peak, they may be discriminated against by using a comparator and noise filter. The short duration noise spikes are removed from the slowly changing signal. If the time constant is chosen to be a few milliseconds, the noise spikes will be disregarded, but only a few milliseconds of the signal will be lost. The variable time integrator was assembled and tested. Before the performance could be fully characterized, it became obvious that the variable time integrator was a complicated solution to a simple problem. 2. Digital Integration If the photocurrent is converted to a voltage and digitized periodically, the sum of the digital representations of the signal is proportional to the integrated signal. This type of integration is inherently more simple because it only involves the digitization step, a procedure necessary even when analog integration is used. The photocurrent was modified by a current-to-voltage converter \which had a drift of less than 0.005% per °C and 0.005% per day 92 (Keithley Model 427, Keithley Instruments Inc., Cleveland OH). The conversion of the current to voltage was the only analog operation performed on the signal, so drifts are minimized. The digital integrator can be made to integrate over a fixed time or peak recognition criteria can also be used. A laboratory minicomputer (PDP Lab 8/e, Digital Equipment Corporation, Maynard, MA) was used to make these operations easier. A real-time clock i I within the computer was used to provide timing pulses for the analog- WWW-37 11" “ " -" .le m 31 hm. to-digital converter (ADC). The converted values were stored and summed. The time-integrated atomic fluorescence was outputted via teletypewriter after each sample. The integrals were stored for statistical treatment, background correction, etc. The variable time integrator was found to produce more precise data than the fixed time integrator for large signals, but when the signal-to-noise ratio (S/N) became small,the computer had trouble recognizing peaks. In practice, the variable-time integrator was used for the fundamental studies, and a fixed time integrator was used for analy- tical purposes. The integrator was tested against the "classical" methods of integrating peaks. A synthetic signal similar in shape to an AF peak was integrated by the digital integrator and also recorded on a strip-chart recorder (Omnigraphic 3000, Houston Instruments Inc., Houston, TX) where the peaks were integrated by planimetry and "cut and weigh" techniques. The synthetic signal was a sine-squared signal, recorded from zero to n, produced by a Wavetek llO signal 93 generator, (Wavetek, San Diego, CA). The results of the integra- tion procedures are summarized in Table I. Table 1. Comparison of Methods of Peak Integration. Relative Standard 5 . Method Mean Area Deviation ._,Tf 5 Digital 0.684 V—sec 0.38% l Planimetry 0.66 V-sec 0.99% t} i' Cut and weight 0.113 g 1.54% ’ Peak height 0.602 V --__ (recorder) Peak height 0.596 V ---- (oscilloscope) The relative standard deviations are based on five samples. When planimetry was used, integration was performed in triplicate and the relative standard deviation represents only the variance among the five samples. The results indicate that the digital integration method is more precise than the other methods. To test the accuracy of the digital integrator, a square wave of known amplitude and duration was used. The absolute accuracy was found to be the same as the precision. 3. Photon Counting The output of the PMT has been assumed to be an analog signal i.e., an electrical current. It is actually a series of discrete pulses. 94 When a photon strikes the photocathode, a photoelectron may be emitted. The probability of this event is given by the quantum ef- ficiency of the photocathode and is wavelength dependent. The photo- electron is accelerated to the first dynode, where 3-4 secondary electrons are produced for each primary electron collected. Each secondary electron is accelerated down the dynode chain (5-l4 dynodes) and finally a pulse containing l05 to 107 electrons appears at the anode. For the PMT's used in these studies, a lP28A or Rl66, the electron gain is approximately l06,so the anodic "current" con- sists of a series of pulses, each containing 106 electrons. The height of each pulse varies because of the statistical nature of secondary emission. At sufficiently low light levels, these in- dividual pulses may be resolved. An integrator may be built on the photon counting principle; the integral is no more than the sum of the PMT anode pulses collected from an AF peak. To perform photon counting experiments, the instrumental system shown in Figure T4 was used. PULSE AMPLIFIER U ( 000 U 6‘ _‘ $5 PHOTOMULTIPLIER RL 00 was DISCRIMINATOR REFERENCE LEVEL Figure l4. Photon Counting Circuit. 95 The current pulses at the anode of the PMT are converted to a voltage by RL' The voltage is amplified by a fast amplifier. The output consists of large pulses from photoelectrons which have been amplified by the full gain of the PMT, and smaller pulses due to electrons which have been generated by the PMT and have not originated from the photocathode. These electrons, which constitute the bulk of the dark current if an analog measurement 'ks performed, arise from cold-field emission, thermonic emission, decay from 19K40 in the glass envelope, cosmic radiation, etc. (89). The discrimin— ator rejects all pulses below a reference level, so it can reject some of the dark pulses. Dark pulses originatingartthe photocathode undergo full amplification and cannot be discriminated from photo- electron pulses. Pulses of amplitude greater than the reference level are shaped for use in the counting circuit and then summed with a digital counting circuit. The circuit of Figure l4 was built with a Motorola MClSlOG (Pioneer Electronics, Detroit, MI) for the amplifier. The amplifier specifications gave the maximum frequency response as 40 MHz. The amplifier was placed on a small circuit board and mounted inside the photomultiplier housing. The circuit was tested by injecting high frequency current pulses at the anode pin of the PMT socket (PMT removed, of course). Under these conditions, the amplifier was found to be usable to 8 MHz (3db point). The amplifier was tested for linearity by preparing a plot of direct current vs count rate. This plot is shown in Figure 15. Note that the relationship appears to be absolutely linear to 1 MHz. 96 .pcmcczu uuwcwo mo cowuuczd m we pczou couoca .mp mczmwd <.szmm:o $10. ~10. 9.0. 0.0. OTC. _ _ a q a OO._uwa04m sf “0 cu 9. 9. 9. .-OBS '1“!!! anOO NOlOHd In '9 ca 52 97 More rapid photon arrival rates produced a larger dc signal, but the count rate actually decreased. This is due to pulse overlap effects at the amplifier. When the arrival rate of photons exceeds a certain value, the amplifier will not have time to respond to individual pulses from the anode. The presentation of Figure l5 is log current vs log count rate in order to compress the data to fit the display format. This also has the effect of compressing errors. Even though the slope is 1.00, at low light levels, as found from a least square fit to the data, the amplifier is nonlinear to the extent of l% for frequencies greater than 100 kHz. This was determined by using neutral density filters in the light path and calculating absorbances. The non-linearity at high photon arrival rates is not really detrimental to the use of photon counting. It can be used at low light levels where photon counting actually provides higher S/N than dc measurements. The improvement in S/N comes from noise due to the statistical nature of emissions from dynodes, secondary emis- sion noise. Since a dynode may produce 2 - 4 electrons for every electron striking the dynodes then the output at the photoanode may differ from one pulse to the next. The dc measurement system responds to the magnitude of the current pulses, and secondary emission noise is included. Photon counting, however, depends only on the presence or absence of pulses, so the signal-to-noise ratio will be improved by a factor proportional to the secondary emission noise. If instrumental bandwidths are the same for both dc and photon counting measurements, the S/N for photon counting is 5-22% better than for dc measurements (97)- V. OPTIMIZATION OF INSTRUMENTAL PARAMETERS The manner in which the experimental variables affect the atomization process must be known if the mechanism of atomization in nonflame cells is to be elucidated. The experimental parameters affect the final readout in several different ways: some parameters actually affect the atomization process, while others affect only the instrumental response to the atomization process. Other param- eters affect both the atomization and the instrumental response. In this section, the influence of experimental variables on non- flame AF is described in detail and results are used to select optimum parameters. A. Optimization of Signal-to-Noise Ratio The instrumental parameters may be varied to achieve several different effects. The most obvious effect is the change in the signal. Most spectroscopists immediately "tune" their instruments for the largest possible signal. Unfortunately the conditions chosen to optimize the instrumental response to a standard may not be optimum conditions for any of the analytical samples. Adjusting the parameters for the largest signal may often result in a situation very prone to noise. If the primary excitation source is indis- criminately adjusted for the largest signal, it may also produce the largest blank, due to scatter. The signal magnitude is not 98 99 as important to the analytical chemist as is the signal-to-noise ratio (S/N), at any given level of sensitivity. The signal-to- noise ratio can be shown to be directly related to the precision of the analysis. The relative standard deviation (RSD), equals the reciprocal of the S/N (98). Optimizing the S/N is a relatively straight-forward task. The experimental parameters are varied until the R30 is minimized; at this point the precision (and the S/N) is the highest. The standard deviations are easily calculated, in fact the computer- controlled digital integrator, described in the previous section, is ideal for a study of this type. The computer can record the data and output a signal-to-noise ratio within a few seconds. This allows the operator to vary the parameters and follow the results as the parameters are varied. If the computer were not “on-line", it could take hours, even days, before the operator would know the optimum conditions for analysis. Signal-to-noise ratio is not the only parameter of interest to the spectroscopist. An optimization for accuracy,sensitivity, or selectivity are alternative goals. The procedure is the same but the S/N is at least as important and as easy to determine experimentally. The low S/N of nonflame atomizers (poor precision) has been said to be its major drawback, so a comprehensive study of the effects of experimental variables on S/N is certainly impor- tant. All the experimental data pertain only to the test element, cadmium. This element was chosen primarily for the ease of atom- ization with the platinum loop atomizer. 100 B. Experimental l. Instrumentation The instrumentation used was described in detail in Chapter IV. A fully automated nonflame spectrometer was constructed by placing the automated sampler and atomizer power control under the direction of a hardware master controller. The integration and readout were under the control of a minicomputer. Since the vari- able time digital integration method is based only on the presence or absence of an AF peak, no timing signals are needed to synchro- nize the atomization and integration commands. A generalized block diagram of the spectrometer appears in Figure l6. The logic and sequencing unit is the combination of a hardware control system and a digital minicomputer. The control system instructs the automatic sampler to deliver a sample to the atomizer. When the sampler completes the delivery step, it responds with a signal to the control system, which then turns the sampler off, and provides a signal which switches elec- trical current through the atomizer. The sample is desolvated, and then atomized when the atomizer is heated. A stream of inert gas carries the atomic vapor to the level of the monochromator entrance slit, where absorption of the radiation from an intense line source produces atomic fluorescence. The fluorescence signal, monitored at right angles to the excitation source, is converted to a propor- tional electrical current by a photomultiplier tube. The elec- trical signal is in the form of a current peak, which is integrated, and ultimately read out as a number. The control system outputs lOl .memEoLpumam mamfimcoz umpmsou=< mo Emgmmwo xoo_m .m_ mczmwd OZUZmDOmm QZ< U50.— e ”.025 .32. . $2522 Azmfisu - C I O 0.. 0 0.0.0.0... . I .A A L“ I D I 0230266.. 5095 M... 3,3,; I..- OO'DOIO 0......' .00.... ICCUIIOI .IOII... .‘.-AAAA“A-AA AAA 1.14 a . a 1:1 .................................. coco-ooooooooo-aoooo.u coauouooa-oooo- o o out...cooIOOQOOIDOHOOQOOH’one-coo".a o .oooouonononsooNQOOHOOONoo...incur-o. o a.ooooooOQoooooooquIHIHo oooooooooooo ooooooooooooooooooooooooooooooooooooooooooooooo o o o o o a o o o o o o o c a a o o o c o no. u o o - ooooooocooooooooooo- 00.006.00.00.o-ooooo-oo-oo o a a u o on.cocoon-oocco-ooooooonouo oooooooooo 102 the data, and then repeats the cycle. a. Control system. The circuit diagram for the control system is shown in Figure 17. The timing sequence is initiated by a pulse from a microswitch on the sampler, which is shaped by NAND gates 1, 2, and 3. The leading edge of the microswitch signal clears all the flip-flops, causing the drive motor on the sampling system to stop. The trailing edge allows flip-flops A and B and NAND gate 5 to function as a single pulse gate, which produces one full clock pulse at the output of NAND gate 5, regardless of the initial state of the clock. The clock pulse is directed to the circuit which heats the loop, so that the atomization time will be as constant and reproducible as is the clock. All the logic level outputs go through protective devices to prevent the logic system from transients. Alternatively, the entire hardware control system may be replaced by a small com- puter (99). b. Fabrication of the loop atomizer. The loop is made from 32 gauge wire, which is 90% Pt and 10% Rh. The platinum-rhodium alloy was chosen for its stiffness and overall strength, higher operating temperature than pure platinum and low susceptibility to oxidation. The diameter of the loop is 2.0 mm, and it is held in place by a spring clip arrangement, which allows replacement in less than a minute. Although fatigue is commonly exhibited in pure metals under repeated heating and cooling, alloying seems to improve the material's fatigue characteristics (100). The platinum-rhodium alloy became weak after repeated use, but not to the extent of 103 .Emumxm Fogpcou memaogpomam we Ewcmmwo uraucwu .u. mczm_. xuaqnu.Illlllllglllllallllux. .xun: IIIIIIIInyuIIIIflIIIIIL. I =11..I| 65... L IUZ?m0~.U.<‘ IJ 1_I|III , 52385.2 50 . Eu II— . so . v. V. - n0 _ _ «0.62 9 T o «00.. 0.. TIII‘IIIIII 104 pure platinum. Loops were generally replaced after several hundred samples were run by which time changes in the loop could generally be seen. c. Solutions. Stock solutions were prepared from pure metals. All solutions were stored in polyethylene, and prepared within hours of use to minimize adsorption on the walls of the container. With the pump-capillary sampling system, the solution to be analyzed was placed in a vessel similar to the mercury reservoir used in polarographic methods. This ensured that the height of the liquid would remain constant to eliminate any siphoning through the peristaltic pump. 2. Procedure The automated sample delivery system had the drawback of pos- sible adsorption onto the tubing. All optimization studies were made using the automated system. However, when working curves were desired, a 10 ul syringe (Unimetrics Universal Corp., Anaheim, CA) was used to place the sample onto the 100p. All of the experimental data presented on optimizing parameters were obtained with the sampling system which delivers the sample from below the loop. A similar optimization procedure with a second sampler, similar to an automated microsyringe, revealed only minor differences in Optimum conditions. C. Experimental Variables The first parameters to be optimized are those which do not affect the atomization process. These parameters include the 105 current to the primary excitation source, the monochromator slit width, the photomultiplier anode-to-cathode supply voltage, and the relative spatial locations of the source,atomizer, and detector. Most of these parameters affect the S/N in a known manner. Increasing PMT supply voltage increases the S/N, so the voltage was continually adjusted upward. The absolute limit of PMT voltage is a function of the photomultiplier; in practice, the photocurrent due to stray light was kept below 10'6 A. Continual use with anodic currents greater than 10"6 can damage the PMT. Adjusting the monochromator slit width affects both the amount of fluorescence and the amount of background which reach the detector. The relationship is different however. Under low background condi- tions, a wide slit maximizes S/N, but if the background is high, a narrow slit maximizes S/N. Hence, with most nonflame atomizers the S/N increases with slit width up to a certain value and then decreases with further slit width increases. The optimum slit width must be determined experimentally. The optical system must also be optimized experimentally. The positioning may be Optimized for the largest signal, as noise is affected only to a small extent, when the obviously improper extreme choices are rejected. The current to the primary excita- tion source (a metal vapor discharge lamp) may be optimized in the same manner. As the current increases, the total source radiance increases, but the radiance over the atomic absorption line may actually decrease. This effect is a result of line broadening in the source at sufficiently large currents. 106 The optimum spectrometer parameters as determined from an experimental optimization procedure are given in Table 2 for AF analyses of Cd. Table 2. Optimum Spectrometer Parameters Parameters Optimum Excitation source Cd metal vapor lamp Source current 1.2 A Atomizer position 2.0 cm below slit Wavelength 228.8 nm Slit width 1.0 mm PMT supply voltage 1000 V 1. Sample size The sample size, loop size, signal, and noise all interact in a complex manner. The S/N is further affected by some of the other parameters (e.g. atomization rate), which also change with sample size. In general, the noise decreased as the loop diameter decreased, and the precision increased. If the diameter were made smaller than 2 mm, the signal began decreasing faster than the noise, so the S/N became smaller. This relationship can be explained as follows: as the loop diameter decreases, the actual point of atomization becomes more readily defined. As the sample is atomized, the metal vapor is swept into the viewing cell by the sheath gas. Although every 107 attempt has been made to ensure uniformity of the sheath, there is probably some degree of turbulence. If the spatial location of atomization is imprecise, the turbulence will multiply this im- precision so that small imprecisions at the atomizer become large imprecisions at the point of observation. As the atomizer size is decreased the variance (imprecision and variance are approximately interchangeable) introduced by the atomizer becomes smaller. At the same time, simply because the atomizer cannot hold a large sample, the signal becomes smaller and some of the other noise sources start to dominate over atomizer imprecision. Other sources of noise are shot noise in the back- ground, shot noise in the signal, and Johnson noise in the elec- tronics. The overall result is that the S/N decreases as the atomizer size decreases beyond a certain size. The optimum atomizer diameter was found to be 2 mm. The optimum sample size was found to be about 4 pl. A complete optimization is not necessary for these parameters. The RSD does not vary by more than a few percent (4% to 8%) as the atomizer size is varied from 1.5 to 2.5 mm, and the sample size is varied from 3 - 8 ul. The relative independence of S/N on small scale variations of instru- mental parameters is advantageous because drifts and small errors will not affect the signal to a large extent. 2. Sheath Gas Flow Rate The sheath gas has two functions. It shields the atomic vapor from the atmosphere and also transports the vapor from the atomizer to the observation zone. Argon was chosen as the sheath gas because 108 of its low quenching cross-section. Actually two parameters must be varied in order to optimize the performance of the sheath. Ob- viously sheath gas flow rate affects the transportation function of the sheath, but the shielding function is affected more by the design of the sheath than by the flow rate of gas. The optimization of the design and construction of a laminar flow head for the gas sheath would be a very time consuming process. It was not attempted in this study. The gas sheath was divided into an outer column to prohibit entrainment of ambient air in the vapor cell anda concentric inner sheath to transport the atomic vapor from the atomizer to the observation area. The flow rates of the inner and outer sheaths were varied and the resultant S/N was recorded. Since the inner and outer sheaths interact with each other, they are plotted on the same graph. Fig- ure 18 shows the results of this study. Contours of constant R50 are drawn for different inner and outer sheath gas flow rates. Note that if the inner flow rate is 2 l/min, then the optimum outer flow rate is about 3 l/min. If the inner flow rate is increased to 3 l/min, now the optimum outer flow rate is a little less than 2 l/min. An explanation for the exact nature of the relationship be- tween S/N and sheath gas flow rate cannot be provided at this time. The reasons for the behavior at some of the extremes are, however, known. When both the inner and outer flow rates are low, the sig- nal is depressed due to the quenching of atomic fluorescence by atmospheric nitrogen and oxygen. As the outer flow rate increases, A Imin. (A OUTER FLOW RATE , .1. N 109 d 7% I l - 1 2 3 INNER FLOW RATE, l/min. Figure 18. Dependence of Precision on Sheath Gas Flow Rate. 4 110 the S/Nimproves (inner flow rate constant) up to a point, then de- grades. This is probably due to turbulence and the entrainment of ambient air in the sheath. If the outer sheath gas flow rate is held constant, the S/N rises to a maximum as the inner sheath gas flow rate increases. If the flow rate increases beyond a certain point, the S/N starts to degrade. The following explanations are plausible: at low inner flow rates the atomic vapor has time to diffuse out of the observa- tion area. This will decrease the S/N by decreasing the signal. At high flow rates the S/N decreases because the residence time of the atomic vapor in the observation zone decreases. Since the integrated signal is directly proportional to the residence time (Equation 28),the signal decreases with decreased residence time. 3. Power Applied to Atomizer Varying the power applied to the atomizer varies both the steady state temperature and the rate at which the temperature approaches the steady state. The power affects the signal to a different extent than the noise. Figure 19 shows a plot of the integrated AF signal as a function of atomizer power. The AF signal clearly decreases as the power to the atomizer increases. This behavior is obviously a residence time effect. Since the integrated AF signal is propor- tional to the product of concentration and residence time, (Equation (28)), then the applied power must be changing one of these param- eters. The most probable mechanism by which the atomizer power affects 111 .LmNWEOp< mzu op umPFaa< cmzoa co Pmcmwm umumgmmucH mo mucmncmama 975;» .mmgon. Avg. 0:0 min .m. ac:m.. 06 ! _ u Lona 'MVBd uaown vauv 112 the signal is by convection. At high applied powers, the atomizer heats the sheath gas just above the loop and increases the transla- tional energy of the gas. This will either cause turbulence and dilute the atomic vapor or simply speed up the gas flow and reduce the effective residence time. The exact means by which atomizer power affects the signal cannot be identified at this time. The noise is also affected by the applied power. Although the signal is probably affected more by the change in steady-state tem- perature (convection currents) than is the noise, the noise is probably affected more by the changing temperature - time behavior than is the signal. At low powers the noise is increased as the sample is atomized slowly and "flicker" effects set in. At high applied power, the atomization process becomes imprecise due to sample explosion from the atomizer. The signal-to-noise ratio will be low (high RSD) at both very high and very low levels of applied power. This is, in fact, the observed relationship as shown in Figure 20. Hence, there is an optimum applied power for maximum S/N. 4. Integration Parameters The computer-controlled integrator used certain criteria to determine if the input is an AF signal or a noise spike. A signal is defined as any input voltage which remains above a reference level for a specified length of time. The latter criterion is similar to the time constant of an analog circuit and will be designated as the "time constant" in future discussion. Both the reference voltage and time constant are set by the 113 .LmNPEou< ms“ o» umwpan< Logo; :0 cowmwumgm we mucmwcmamo .om mg=m_. 323 .539. ON o.m 0.0 06 fi - . — O. 0. ON mm °/o‘ NOILVIASO OBVONVLS HAHN-138 114 computer after an initial dialogue with the operator. Increasing the reference voltage and the time constant tends to discriminate against the small, short noise spikes. The amplitude distribution of noise spikes is probably Gaussian. If the reference voltage is adjusted to discriminate against 99.7% of the noise spikes (30), for example, then increasing the reference voltage beyond this would only affect the signal. The time constant probably affects the noise in a similar manner. At very large time constants and reference voltages a portion of the signal is truncated. In fact, the computer may completely miss some signals. This problem does not occur except when the signal approaches the limit of detection. The signal begins to resemble noise, and the computer may not be able to tell the difference. A plot of the effects of time constant and reference voltage on precision appears in Figure 21. Again contours of constant RSO are plotted for various reference voltages and time constants. If the time constant (or reference voltage) is held constant at any reasonable value, then the S/N is seen to go through a maximum as the reference voltage (or time constant) is varied. The experi- mental relationship tends to reinforce the strength of the predic- tions. 0. Interaction of Variables If the variables were independent, the entire optimization procedure would be easy. Unfortunately this is seldom the case in any experiment, and is certainly not the case in this experiment. 115 OF .mcmumsmcma LoumcmmucH co cowm_omca we mucmvcmamo 9.23 .Omm 3; ....Z<...w200 mi... N a m n 0 n v m ._N scam.. p a q . q 1_\ — Av\eoom Mu 3 mod n... mm v0.0 N 0 8.0 M . 0 moo .1 w. 2.0 0 .3 A 116 The experimental variables clearly interact with each other. If the sheath gas flow rate is increased, the atomizer temperature de- creases due to the cooling effect. This in turn changes the atomiza- tion rate which in turn changes the temporal characteristics of the peak. The integrator parameters (reference voltage and time constant) may no longer be optimized. The primary effects of changing the sheath gas flow rate are changing in shielding efficiency and transport rate of the gas sheath. Under certain conditions, the primary effects may not be much larger than the secondary effects, so optimization becomes quite complex. Some of the interactions of the experimental variables are presented in Table 3. This is by no means complete; many other effects are present. The table does, however, illustrate the major interactions and possible mechanisms for the interactions. The data presented in Figures 18-21 have been produced by vary- ing one parameter at a time and recording the S/N while holding all other parameters constant. The exact optimum value is dependent on the other parameters, although the general shape of the plot of S/N vs variable is not seriously affected. After initial optima were chosen, each parameter was reoptimized while holding the other parameters at their optimum value. The results of the final opti- mization are presented in Table 4. The agreement between the values of the final optimization and the optimum values of Figures 18-21 indicate that the conditions under which the data were obtained were close to Optimum. If good initial estimates cannot be made, then the optimization procedure 117 awe venom: up acmamcoo we?» use wmepPo> mucmcm0mg 3m: om mmmcmco xmma m< xmwa xmwa m< .< 00 mamcm mmmcmcu mo mamcm mmmcmgu yo copumcav we?» use muzpwcmme use mcmpmsmgmm cowumcmmuc. mFaEmm mNFLonm> ow Lmzoq cmgmpz mmcvzamm Lm~weoum Poou mmumc 30.0 cow: gmzoa amppaa< zuwuo_m> mcwupmwcm 30—0 cw wmcwgo opmaamuo Low umumm: mocmpancau mmmamo mung 30F. mum; sop» cw mmmcagu mczumcmasmp Locum: mam cummgm mNPm mFQEmm Luzon now-aa< mama 30’. wow cummgm umuum$$< LmumEeme umem> LmumEmgmm mcmpmsmgma .mucmsagumc. 0o cowuumcmucH .m mpnmh 118 Table 4. Optimum Platinum Loop Atomizer Parameters Parameter Optimum Sheath gas Argon Inner sheath flow rate 2.0 l/min. Outer sheath flow rate 2.0 l/min. Atomizer current 3.6 A Atomizer voltage 1.47 V Sample size 4 pl Integrator reference voltage 0.08 V Integrator time constant 8 becomes extremely complex. Several local maxima exist for the varia- tion of S/N with the experimental variables, and care must be taken to ensure that the optimization procedure does not stop at the local maximum, but instead searches for the unique global maximum. It should be noted that the optimization study reported here required approximately 20,000 AF analyses of cadmium. Only the extent of automation and computer control allowed this study to be completed in a reasonable length of time. The large number of data points, however, diminished the possibility of the optimization being incomplete. 119 E. Analytical Results 1. Stability After the optimum parameters were chosen, a study of the long- term stability of the nonflame AF system was performed. The results of the study are shown in Figure 22. One hundred separate samples, each containing 2x10.8 9 of cadmium, were analyzed by the fully optimized automated nonflame spectrometer. The analysis took one hour, and was performed entirely under the direction of a laboratory minicomputer. No operator adjustments were made during the duration of the analyses. The relative standard deviation of the 100 samples was 9.1%. The instrument exhibits no long-term drift, but the results are less precise near the end of the analysis. This behavior points out the advantage of recording signal-to-noise ratios rather than signals. The signal remains constant, but the noise increases with time. The increase in noise is due to several factors. The atomizer properties for example, change over long periods of time. Repeti~ tive heating and cooling changes the metal's characteristics. After hundreds of samples, the loop size has generally changed from the starting size. The atomizer becomes weaker and starts to sag, so the overall shape of the atomizer changes with time. The sampling precision also changes with time. The tubing in the peristaltic pump will lose resiliance as it is continuously compressed. Unless the loss is constant, a variance term will be introduced by the sampler. The precision early in the analysis is excellent. The RSD 120 100.1 . 80"I . o . 70- o ' 60" '0: 50-1 0 . 404 . ’ SAMPLE NUMBER 30- ', - I U U I I l ~3'0 -2O -10 0 10 20 3 DEVIATION FROM MEAN. % Figure 22. Long Term Stability of Nonflame Spectrometer Cadmium Under Fully Optimized Conditions. 121 of the first ten samples is less than 2%. The imprecision is close to random: 62% of the samples are within one standard deviation from the mean; 89% within :20; 100% within :30. The distribution would be 68%; 95%; and 99.7% if the errors were truly random. 2. Calibration Plot a. Photocurrent integration. The analyte concentration was varied,and the integrated signal was plotted as a function of con— centration in Figure 23. The data are presented on logarithmic axes to illustrate the length of linearity of the plot. The standards ’10 g to 2x10-8 varied from 10 9 if cadmium. This corresponds to con- centrations in the range of 0.04 to 2 ug/ml. The uncertainty in the experimental results is indicated by the length of the vertical bars on the signal axis. The relative standard deviations were approximately 9 to 7% and independent of the signal. This tends to indicate that the dominant noise source is flicker in the primary excitation source. Both the precision and minimum detectible concentration can be improved by using a light source more stable and more intense than the metal vapor discharge lamp. An electrodeless discharge lamp, for example, should sub- stantially improve precision. The slope of the log-log plot was determined to be unity within experimental errors. The slope was found by a least square fit which assumed the error in the signal to be much larger than the error in concentration. This assumption is probably better than any of the assumptions made in Chapter III. b. Peak height measurements. The time duration of the AF Figure 23. Integrated Fluorescence as a Function of Cadmium Con- centration Under Fully Optimized Conditions. .. 100 122 O F '— 4 ! ‘OIXO EONHOSEIHOD‘H GHLVHSEIINI l 8.0 20 2.0 4.0 0.8 0.4 CADMIUM PRESENT, ng 0.1 123 signal varied from 0.08 sec to approximately 2 seconds over the con- centration range studied. When peak heights were plotted against concentration, the calibration plot was markedly curved at high con- centrations. These data are presented in Figure 24. The curvature is not bad enough so that the calibration plot can be considered unusable, but precision is generally less in curved regions of analy- tical working curves (101). The signal at the limit of detection is presented in Figure 25. The signal can readily be seen above the baseline. If the noise can be considered to be 1/5 the peak-to-peak excursion of the back- ground (93), then the S/N is approximately 20 for these signals. The computer-controlled integrator cannot discern the signal from the noise based on the reference voltage and time constant which have been given to it. At low signal levels it becomes advantageous to switch to a fixed time integration method. c. Photon Counting. The sensitive integration method is used to perform photon counting measurements. The photon counting system shown in Figure 14 was used. The discriminator was the discriminator of the EU-805 Universal Digital Instrument. The UDI was modified by replacing the standard input transistors with high speed, low noise transistors and by adding a ten-turn potentiometer to the internal discriminator. The photon counting was performed for a 10 second period encompassing the moment of atomization. The results are shown in Figure 26. .mee.e.eeeo een_e.peo »_.=L Lace: cowpmcpcmocou SawEumu mo cowuuczm m we mucmummcoafie ow50p< .mme .em mL:m_. mo_xo .Ezmmmme 22:53 Qm oé o.~ md to _.o _ d1 d _ _ _ 124 l 9 301x11 'LNBHHRDOLOHd )IVBd 1 00. 125 >wu\umm om .mswh umwx< Pmuco~_coz .mp_:: acmcuwng< .mucmommgozpm um_x< Peowpgm> .ee.eedede Le SFEPS nee Be Feemem wee .mm deemee 411.111.111.11 Ears mu m . u o_-o. op 0m 0m 126 .:o_pecycmocoo Suwanee we cowuucae m we mama ucaou :ouoca .om mcsz. a: .thmwma .23..>_O<0 8.. 9. 3. to oo.— u MAM—04w d m o— n I. 0 N m N 1 OF .M «a. S n4. 3 I In 9 127 The calibration plot is linear for a shorter concentration range and exhibits non-unity slope. The most probable reason for the lack of agreement with theory is pulse pile up effects in the photon counting system. If 104 counts are accumulated from a signal only 80 msec wide, the peak count rate would have to be greater than 250 kHz. Count rates above 10 kHz begin to give noticeable nonlinearity, so the experimental data cannot be expected to be linear. If a faster counter or one with mathematical correction for pulse overlap is used (102),then photon counting becomes the method of choice with low back- ground nonflame atomizers. VI. CHARACTERIZATION OF THE ATOMIZATION PROCESS The events leading to the atomization of the sample significantly affect the atomization process. If these events are known and under- stood, they can conceivably be controlled. In Chapter V, it was shown that if the instrumental parameters are controllable, they may be optimized so that the best precision may be obtained. If the atomiza- tion process can be controlled, then it may also be optimized. Both the events leading to atomization and the events following atomiza- tion must be elucidated if the overall atomization process is to be understood. A. Characterization of the Platinum Filament Atomizer The repetitive sequence of heating the filament and then cooling it (by placing the sample on it) will cause changes in both the chemical and physical nature Of the filament material. The constant heating and quenching cause the filament to become brittle, to lose strength, and to change electrical resistance. The latter effect is due to evaporation of filament material at the atomization temperature. 1. Evaporation of the Filament Material The rate of evaporation of material from a filament is well known and may determine the useful life of the filament (103). The evaporation rate is known to be higher when alternating current is 128 129 used than when the same rms direct current is used. This is due to the fact that the rate of evaporation is an exponential function of temperature (103). If temperature varies sinusoidally with time, as in the case of a filament heated by AC, the average rate of evapora- tion over one period is greater than the rate of atomization at the mean temperature. For a tungsten filament heated to 2,000 K with an alternating current, the temperature varies over a 157 K range,and the evaporation rate is 9.53 times the rate calculated for the same direct current (104). The evaporation will affect the platinum loop atomizer by decreas- ing the diameter of the filament and increasing the resistance. The steady-state temperature of the loop will then decrease if the applied voltage is kept constant. The extent of the evaporation of tungsten filaments is small; only about 0.02% per hour (104). Platinum, how— ever, is considerably more volatile than tungsten, so the evaporation rate should be higher than for tungsten. Since no data are avail- able which describe evaporation from a platinum filament,the evapora- tion losses were measured experimentally. Unfortunately the vari- ance in weighing the filament (total weight of about 20 mg) was about 0.1 mg; the measured loss due to the evaporation after about 1,000 analyses was approximately 0.1 mg, so the results are inconclusive. Even if the variance were disregarded, an evaporation loss of 0.1 mg would change the electrical resistance only by 0.5%,and the steady state temperature would decrease by about 6 K. This change is prob- ably insignificant to the experiment. 130 2. Changes in the Physicochemical Properties of the Atomizer After several hundred analyses the properties of the platinum loop atomizer change enough to be readily detected. The loop becomes brittle and is easily broken. The tensile strength is decreased and there is little spring—type action from a coil. Since the changes in the loop are due to atomizing a series of samples, the changes must be analyzed as possibly being competitive reactions to atomization processes. a. Chemical composition. A platinum loop used for approximately 450 analyses of 2 ug/ml cadmium solutions was analyzed to see if any cadmium has been leached into the platinum. The loop was taped directly to an aluminum x-ray target, and the x-ray fluorescence spectrum was scanned. A Philips vacuum x-ray fluorescence spectrometer was used in conjunction with an ethylenediamine d-tartrate (EDDT) analyzer crystal. The spectrum showed no lines other than platinum. An ex- panded scale scan over the angle where the Cd K 1 fluorescence would 8 be expected showed only platinum lines. An extraction was performed to eliminate the bulk of the matrix. The filament was dissolved in aqua regia and then the solution was neutralized with sodium carbonate. The pH was adjusted to 12 and then a solvent extraction was performed with a 0.1% dithizone solu- tion (w/w in CHC13). The dithizone quantitatively extracts any cadmium from a pH 12 aqueous solution. Drops of the dithizone extract were placed on an x-ray target and the solvent was evaporated with the aid of a heat lamp. The evaporation was repeated until the entire chloro- form layer had been evaporated onto the target. 131 The analysis showed only a little platinum and traces of lead. The lead was probably a contaminant in the acid used to dissolve the filament. There was still no trace of cadmium. The limit of detection for cadmium was estimated to be 1 pg. It is obvious that any change in the properties of the platinum were not due to the dis- solution of cadmium into the platinum. b. Physical properties. The surface of the loop was examined with an optical microscope. New loops were homogeneous and shiny whereas old loops showed occlusions. These appeared as white granules, 10 to 100 m in diameter, embedded in a gray matrix that appeared to 'be powdery. Since chemical analysis showed no major constituents other than platinum, these occlusions were thought to be various forms of platinum. A microtome capable of sectioning the material could not be found, so only the surface layer was examined. Platinum is known to have several crystal structures (105). The occlusions may easily be platinum of a different crystalline form. The gray powdery like base is probably spongy platinum. This is the natural form of pure platinum prior to the annealing and ex- truding of the platinum into a thin wire. c. The platinum-rhodium alloy. Some of the physical properties of the atomizer are improved if an alloy is used rather than the pure material. A platinum-rhodium alloy (90% Pt - 10% Rh) was found to be superior to pure platinum. The alloy has increased tensile strength, hardness, and has a higher melting point when compared to pure plat- inum (100). The differences in these parameters are minor for new 100ps, although the alloy appears to be stronger and more flexible. 132 The superiority of the alloy is more evident after many analyses. The Pt-Rh alloy stays flexible and strong for many more analyses than does pure platinum. The platinum loops could be used for several hundred analyses, but the Pt-Rh loops could be used several thousand times. The continual heating and cooling affected the strength of the atomizer more than any of the other physical characteristics. The atomizers started to sag and if they were deformed slightly, they would not spring back to shape.but several thousand analyses were needed to see this effect in the Pt-Rh alloy. All data obtained with the "platinum" loop atomizer were actually obtained on a Pt-Rh loop atomizer. The loop was changed every 500 samples, so aging effects were minimal. B. Events at the Atomizer Surface 1. Desolvation Several important steps must occur before atomization. First, a sample must be placed on the atomizer. As the atomizer is heated, the solvent will be driven off. During this step, the atomizer temperature will be fixed at the boiling point of the solvent if the applied power is chosen judiciously. If the heat gain by electrical heating is so high that the heat loss due to the latent heat of vaporiz- ation of the solvent, plus conductive,convection, and radiative losses, does not keep the temperature reasonably constant throughout the desolvation step, then other effects set in. The sample can explode off the loop as it is heated, for example. The precision is rather 133 poor under these conditions. After the sample is desolvated, it will be in the form of hydrated salt particles at the surface of the loop. The loop temperature then rises toward the steady-state value. As the temperature increases, the particles lose water of hydration. It is improbable that the heat loss in this step affects the loop temperature because the par- ticles contribute about 2 ppm to the mass of the atomizer. The atomizer temperature continues to rise and as it does, the particles at the surface may be changing crystalline form, or even chemical composition. The analyte salt will be in whatever form is thermodynamically favored, and the thermodynamics are obviously tem- perature dependent. All the reported experimental analyses done in this lab have been performed with cadmium in an aqueous chloride matrix. Cadmium chloride, CaClz, does not decompose or form other compounds when heated. 2. Atomization The anhydrous particle is ultimately vaporized. Since atomic absorption and atomic fluorescence clearly show the presence of cadmium atoms, the molecular CdCl2 must be atomized. Unfortunately, the exact mechanism of atomization cannot be clearly identified at this time. When CdCl2 is heated, it simply boils. One would expect that an equilibrium is achieved at the boiling point, 1233 K. The molecular vapor is not heated in the volume directly above the atomizer, in fact, it is cooled, so dissociation of molecular vapor is an improbable atomization mechanism. The atom production must then take place at the atomizer surface. In other words, thermodynamic considerations 134 must make the reaction CdCl2 + Cd + 2C1 possible at a heated platinum surface. 3. Time Resolution of Events Prior to Atomization It is possible to study the desolvation process by recording the sample size as a function of time. This has been done only in a qualitative fashion, and the results show that even desolvation at a platinum surface is a complex process. The history of a sample droplet on a platinum loop atomizer was recorded with high speed motion picture film. A Pathe camera body was used with a f/35 12-120 mm Angenieux zoom lens plus a series of macro lenses. The lens combination permitted 1:1 reproduction ratios. The film was Kodak 4-X Reversal Film (type 7277). The camera speed was adjusted to 80 fps and calibrated by photographing a digital clock. A sturdy tripod was used to support the camera throughout the study. The enlargement necessary to photograph adequately the events at a loop only 2 mm in diameter will adversely affect the depth of field. The aperture had to be decreased so that the entire 2 mm diameter of the loop would be in focus. At f/22, the aperture used in the experiment, the depth of the field was less than 5 mm. The small aperture obviously limits the light throughout and hence the S/N of the final print. The film had an ASA speed of 800, and even then, light sources using over 500 W had to be placed in close proximity to the atomizer to provide the proper exposure. 135 The last three paragraphs simply say that the images are neces- sarily fuzzy and grainy. The photographs are presented in Figure 27. The first frame in the series, Figure 27A, shows the sample drop- let on the loop. The electrical power has been turned on, but the drop- let has not been noticeably affected. The top frame is considered time zero for the following discussion. The second photograph, Figure 27B, shows the droplet 1.44 seconds later. As it is heated, it starts to boil and expand. The droplet is originally in contact with the entire loop, but eventually looses total contact and becomes attached at only one point. This can be clearly seen in the sequence shown in Figure 27B. The frames are 12.5 msec apart; the bottom frame occurs the latest in time. Figure 27C shows the situation after approximately 5 seconds. The droplet is becoming smaller and smaller. It is clearly migrating toward one spot on the loop rather than evenly coating the loop with salt particles. This is an important point that will be referred to in some of the following discussion. The point to which sample migrates is the coldest spot on the loop. It is also the lowest spot, because it is the closest to the source of the argon sheath. The top frame of Figure 270 shows the last step of the desolvation. This takes place 7.14 sec after time zero. The middle frame, 12.5 msec later, shows that the result of the desolvation is clearly a clump of salt particles, rather than an even film. The bottom frame shows the atomization. The salt is gone and a vapor cloud can be seen below the atomizer. 36 1 .LoNPEou< goo. Eocwuo_o ocu pm mpcm>m mo co?papommm wswh .NN mesmw. uom ¢_.n .a umm m .u umm ¢¢.— .m o mEmH .< 137 Figure 27E, at time 11.20 sec, shows the loop glowing. The atom- ization is complete and the loop has reached a steady-state tempera- ture of approximately 1500 K. 4. Post Atomization Events The events occurring after the atomization were monitored by following the atomic concentration as a function of time. The atomic concentration (as measured by AF) vs time profiles were recorded at various heights and various lateral positions above the loop. The data appear in Table 5. The data cannot be represented adequately by a graphical presen- tation. There are no good ways to plot four parameters (AF radiance, time, lateral, and vertical position) as a function of each other. Some general comments about the data must precede a grossly inadequate attempt at presenting the data. The time axis in the table has been arbitrarily set to zero just before the appearance of the AF signal. The lack of knowledge of the precise moment of atomization makes it impossible to reference the time axis against a standard. The inert gas sheath took the form of a column about 1 cm in diameter. This size had been found to be Optimum in studies performed by Mr. Akbar Montaser in this laboratory (7). He chose this design on S/N criteria for integrated AF peaks. The optimal sheath design is detrimental to the spatial studies be- cause sizable atomic concentrations were found outside the sheath. The sheath, in fact acted to create a toroidal atom population density as it tended to move atoms in the center of the atomic vapor cluster more quickly than atoms near the edge. 138 The viewing area was restricted to a sphere approximately 5 mm2 in area. The sheath gas source was 38 mm below the loop. This was the maximum distance that could be arranged. Hopefully the sheath gas would spread out in a laminar flow at this distance above the source. Several interesting trends can be seen in the data of Table 5. First, when the observations are made directly above the loop, the. atom population is spread out over the entire lateral measurement range. Second, the atomic concentration near the atomizer is still appreciable, although certainly decreasing, even seconds after atomization. These findings clearly show that the environment near the atomizer is very turbulent. This could be due to the atomizer itself, which obstructs the sheath gas flow, or due to thermal currents introduced by the high temperature of the atomizer, or due to rapid diffusion by the heated atomic vapor. The time duration of the AF peak is shorter at heights well above the atomizer than right at the atomizer. The atoms clearly exhibit large lateral diffusion effects. With the observation height just above the atomizer, there is still appreciable concentrations of atomic vapor at the extreme horizontal positions. The concentration at the sides, for low observation heights is about the same as for high heights, but the concentration at the center is higher at points well above the atomizer. The atom cloud seems to have spread to its maximum width just above the atomizer. An overall picture is difficult to obtain. The sheath gas affects the measurement (by allowing quenching collisions) as well as by chang- ing the shape of the atom cloud. The results, which are tentative Time, sec. Peak Height (Ax109) 1.8 2.4 3 3.6 4.2 4.8 5.4 6.0 Photocurrent (Ax109) at Time T, sec. Spatial and Temporal Atom Distribution as a Function of Time 1.2 Table 5. 0.6 O Lateral Position Height 0 mm above loop OONNLO NNNNN F-d'tor—Q' 0000000000 OQONN OOOOO OQOQQ? POP—CO ONl—OO F-OF-F-f— F-r—MNM O O O I <- mm¢Nl\. F-OOOF" OOOOO LOCI-DO MNOQ’ OOtOt—F- Table 5 - Continued Lateral Peak 51 Position Time, sec Photocurrent (Ax109) at Time T, sec. 1.2 Height 4.2mm above loop Heigh (AxlO 6.0 5.4 4.8 4.2 3.6 2.4 1.8 0.6 O kaoan F-r—r—F-F- fivmmw “5?me mmmmm OOOOO mmmmm OOOOO VQL'VVQ: OOOOO LOLOQ'Q’Q‘ OOOOO [\CDLONN OOOOO ONOr-r— f-f—‘O-‘F-l— r—MNOO NNNNN LOP-0mm Q'Q'Q'VQ' ONNO‘F— Q'MQ’Q'LD V'Q'mhxfi 00000 00000 momo 00506: OOOr—F: 5.6 mm 140 above loop \OKDO‘wN CDNLOOOD. OS'U'Q'MO'; NNNNN OOOOO NNNNN OOOOO NNNNN OOOOO mmmmm OOOOO LOLDLOLD'Q.’ OOOOO oomoztx F:F':F_'OO 030001fo F-(NICCIP-N Gmmxooo VMVMM Loom“)? MVNNN VONI'EF— can—.ooo 00000 0.35 0.70 1.05 1.40 7.0 mm above loop FNQ’MM LONwON MQ'MVM 0.1 N!— 00 0.1 NNr-NN OOOOO VMQ'Q'Q' OOOOO NONww OOOOO tooounm r—NNNN LOONQDIN m-a-Nmm VOO‘ON OOOOO LOOLD MNOV OOOI—F: Table 5 - Continued 8.4 mm above loop LON!— O O 0 ”PF ova-co MMQ‘ PF 00 NN COO mm:— 000 VQ'N COO ONO r—OO LOCO MNN cad-m NM? gm> mgp Co 8.1 000m 009 com I._.n:>> HZm com 09 comwgmasou on _ _ _ Pmuwumgoogk Avocumz mucmncomn< mcwu_ew4v Poucwswgmnxu u . _ .om mc=m_a nd 0.— m; 0N n.“ HONVBHOSSV ONLLIINH 153 Table6 Experimental Conditions for the Determination of Stray Radiant Energy A. Instrumentation Hollow Cathode Discharge Lamp JA-45462 (Cd), Fisher Scientific, Waltham, MA. Hollow Cathode Power Supply EU-703-30, Heath Co., Benton Harbor, MI Monochromator EU-700, Heath Co., Benton Harbor, MI Burner "Tri-Flame," Fisher Scientific, Waltham, MA. Photomultiplier lP28A, RCA Electronic Components, Harrison, NJ. Photomultiplier Power Supply EU-701-30, Heath Co., Benton Harbor, MI Current to Voltage Converter Model 427, Keithley Instrument Co., Cleveland, OH. Readout EU-805 Universal Digital Instru- ment, Heath Co., Benton Harbor, MI. B. Parameters Hollow Cathode Current 10 mA Wavelength 228.8 nm Photomultiplier Supply Voltage -l000 V Integration Period 10 sec. Slit Widths 40 to 2000 pm. 154 single deviation of any measured absorbance from the calculated ab- sorbance is l7%, and the mean deviation is l0.4%. The good agree- ment between results obtained with the varying slit width technique and the measured limiting absorbances is an indication of the general validity of the method. The varying slit width technique should be very useful to the atomic spectroscopist for obtaining SRE levels in monochromators under actual experimental conditions used in analytical procedures. The % SRE is easily obtained from a plot of photocurrent vs. slit width. Measurements are made under high signal-to-noise ratio condi- tions, and therefore measurement precision is generally quite high. 2. Nondispersive Optical Systems. a. Advantages. For many AF measurements, the transmission of the monochromator will limit the maximum S/N. Since the atomic fluores- cence is already highly monochromatic, perhaps a monochromator isn't really needed. Jenkins (ll7) proposed that filters be used. Walsh suggested that a solar blind PMT could be used without filters (ll8). This type of PMT has a photocathodic surface which is insensitive to radiation of wavelengths longer than 320 nm. Such an instrument was built and characterized by Vickers et al. (ll9). An improvement of 700-fold was found for the AF signal from zinc when either a filter, or solar blind PMT was used in place of a dispersive monochromator. b. Experimental. Atomic fluorescence data were obtained while using a monochromator, filter and just a solar blind PMT. The same PMT was used throughout the experiment. The filter-solar blind combination was found to be superior, probably because the background 155 observed by the solar blind was high, and degraded the S/N. The background signal when just the solar blind PMT was used was l0'4 A. This is largely due to scattering of the light source radiation (including nonresonance lines) from the reflective surfaces in the atomizer assembly. The background signal with the monochromator was minimal, prob- ably because the monochromator has well defined spatial resolution as well as wavelength resolution. One of the reasons that the back- ground was so high when the solar blind PMT was used is that the solar blind "sees" a great deal more than just the atomic vapor. The monochromator also produced the smallest signal. The filter was a compromise; it excluded the background to a far greater extent than the signal, thus increasing the S/N. The filter cannot be considered to be optimum. A much better method would be to limit the aperature of the solar blind PMT and/or design the atomizer chamber to contribute minimal scattering to the AF signal. B. Photocurrent Integration The photocurrent was integrated by a computer-controlled fixed- time digital integrator. The software was kindly provided by Akbar Montaser (7). The computer was also used to supply timing signals to the sampler and to provide control over the electrical heating of the atomizer. The fixed-time integrator summed a number of analog-to-digital conversions over the atomization time. The atomization time was actually set to be the product of the number of digitizations and 156 the digitization period. The latter two parameters are set in an initial dialogue between the operator and the computer. The integrated AF signal is printed out on a teletype in ar- bitrary units; these have been converted to coulombs for presenta- tion in this work. C. The Sampling System The sampling system described in Chapter IV was replaced by an adaptation of a commercial system. A micro-dispenser, capable of dispensing 0.25 to 20 ul samples, (Model ULD-020, Hacker Machine, Dansville, MI) was used in conjunction with a syringe needle. To place the sample on the loop, the syringe needle was lowered to the atomizer and then the dispenser pumped a sample onto the atomizer. The syringe needle was withdrawn after the sample had been placed on the atomizer. The position of the syringe needle was determined by pneumatic cylinders. The sampler needed only one pulse to complete the entire sampling sequence; the timing was performed under computer control. D. Temperature Programming One of the largest limitations of the platinum loop system is inherent in the atomizer power control. The power setting is critical; the applied power must be low enough to gently desolvate the sample yet high enough to atomize it. Choosing the lower level is difficult, but not impossible for cadmium samples in aqueous matrices. It might, however, be truly impossible for other elements or other matrices. 157 The obvious solution is to use a multi-step heating sequence. A low power may be used to desolvate the sample, followed by a higher power to ash the sample, and a still higher power to atomize the sample. The power applied in each step, and the length of time that the power is applied, must be easily varied. A flexible, programmable power supply was fabricated by control- ling a commercial dc supply (LK 350, Lambda Electronics, Melville, L.I., NY) by a computer. The computer was programmed to use a digital- to-analog converter (DAC) to produce any desired voltage in the range 0 — l0 V. This voltage was buffered and presented to the power supply. The power supply followed the input voltage, and was capable of con- trolling up to 35 A. The precise voltage applied to the atomizer, and the length of time this voltage would be applied, were established in an interactive dialogue between the computer and operator. The computer program was written by Mr. Akbar Montaser (7) and used with his kind consent. The optimization procedure is actually easier with a three step program than without it. The temperature during the first step is adjusted slightly higher than the boiling point of the solvent. The temperature in the second step is adjusted so that the sample is heated hot enough to vaporize any organic material present, but not hot enough to vaporize the sample. The difference in the ashing and atomizing temperatures is usually 1000 K, so this step is not usually critical. The atomization step is also relatively noncritical. The integrated AA or AF signal was shown to be independent of atomiza- tion time by Equation (28). 158 The time for each of the steps in the temperature program is easily optimized. The step should be terminated as soon as the de- solvation (or ashing or atomization) has been accomplished. If the time selected is much longer than necessary, the lifetime of the loop may be affected adversely. There will be no effect on the signal. The temperature of the atomization step can be expected to pro- duce results similar to those in Figure 19 and 20. The effects of varying the ashing temperature are shown in Figure 3l. The matrix, 6% bovine albumin, approximates a serum matrix. This matrix will be referred to as "serum" in the following discussion. The serum has approximately correct amounts of sodium, potassium, etc. added to the protein. The ashing of a viscous, high organic content matrix would seem, at first glance, to be critical. In actuality, there is a wide range of temperatures which give acceptable results. The S/N does go through a maximum, however. The length of the vertical bars in Figure 3l represent the relative standard deviations of three measurements. The best precision (highest S/N) occurs at the same location as the peak signal. This is to be expected, because the ashing step will affect the signal far more than it will affect any noise source. Note that the temperature axis in Figure 31 has no units. The "temperature" programmer actually controlled the voltage across the filament, rather than the temperature of the filament. Assuming only Ohmic heating, then the temperature T is given by 2 _ V 159 .Eszam COP F e? no _E\m3 _ .mgsumgmaemh mcwcm< co mucmummgoz~d umumgmmch yo wucwucmamo mtz: >mmm cw E:?Evmu Low po_a cowumcnmeu .Nm mgzmwm EBSzoEEzmoon 22.220 0. i 0; So _oAV :11 m . mwmv_ Pumz Eacom maomaa< 12 Q o.'aowaosauon13 OBLVHOBLNI 163 \ J “<32 mn_0._m >t23 164 explain these data. The data from the AF of cadmium in the serum matrix are linear, but show log-log slopes that are less than unity. The same holds for the NaCl matrix. The possibility of condensation reactions in the vapor phase, as proposed by West (48), or quenching due to col- lisions cannot be clearly distinguished by this experiment. The atomization for cadmium at a platinum loop is high1y efficient for aqueous solutions, but may be easily dependent on the matrix. The interference mechanism cannot, as yet, be determined exactly. F. The Graphite Braid Atomizer In separate work in these laboratories Akbar Montaser (7) in- vestigated the possibility of using a graphite braid as a nonflame atomizer. His initial investigations were quite promising, and a collaborative study ensued. The results of the collaborative study can be found elsewhere (8). The calibration plots for the integrated AF signals from cadmium and zinc were linear over 2-3 orders of magnitude and detection limits were approximately 10"] 9. Copper and lead were measured by AAS and showed similar linearity and detec- tion limits. These results were obtained without a rigorous optimiza- tion. The quality of the initial results indicate that the graphite braid atomizer (GBA) warrants further study. l. Advantages The atomizer is composed of thousands of fine graphite fibers woven into a braid about l.5 mm in diameter. The braid is strong, 165 flexible, and can readily be heated to 2900 K with less than 300 W of applied power. The GBA has several interesting properties which contribute to its utility as an atomizer. a. Desolvation. The largest difference between the GBA and other carbon or graphite atomizers is the ease with which samples may penetrate the GBA. As soon as a sample is placed onto the atomizer, it soaks into the braid. After the sample has soaked into the braid (less than l second after sampling) the spatial distribution is estab- lished and is independent of time. Varying the time between sampling and desolvation affects neither the mean nor standard deviation of the observed AA or AF signal (8). Prior to the analysis, the sample resides in the volume between the thousands of graphite fibers. When heated, these fibers act to surround the sample with a highly reducing atmosphere. The pos- sibility of sample explosion is minimized by the micro-furnace en- vironment. The heating during the desolvation and ashing steps is much more efficient due to the large surface area of the GBA. b. Atomization. The porous nature of the atomizer also makes significant changes in the atomization process. After the sample is atomized, the atoms remain in contact with the braid atomizer for a much longer time thanikn~other filament atomizers. This obviously affords more complete atomization and should give more freedom from inter- element effects. The atomic vapor must diffuse through the atomizer, and the diffusion rate will be different for vapors of different atomic weights. This effect, however, cannot be measured at this time. 166 2. Optimization of the Graphite Braid Atomizer The same general procedure presented in Chapter IV was repeated for the GBA. The results will not be presented in detail, but ref- erence will be made to some of the more important features. The argon sheath gas flow rate, sample size, and position of the viewing window were optimized for the largest signal-to-noise ratio. The results were quite similar to those obtained with the platinum loop atomizer. The temperature vs power relationship was characterized. Tem- peratures were measured with an optical pyrometer. Only 50 W were needed to heat the atomizer to 2200 K. The relationship between applied power and optical temperature was approximately linear over the region llOO to 2200 K. The power vs temperature relationship was not measured for temperatures less than llOO K due to the lack of an acceptable temperature transducer. The effects of atomization temperature on AF signal are pre- sented in Figure 33. The temperature axis is in terms of voltage, the controlled parameter. The vertical bars represent the relative standard deviation of 3 samples. The best precision occurs when the applied voltage is 3.5 V; the maximum signal occurs when 3 V are applied across the atomizer. Thus the optimum S/N occurs when the atom- izer is hotter than needed for maximum signal, whereas the opposite was true for the platinum loop atomizer. The micro-furnace environ- ment of the braid allows greater heating without sample explosion. This can be expected to minimize matrix effects. Figure 33. Dependence of Integrated Fluorescence on Atomization Temperature of GBA. 167 mtz: >m._._z: '9 to o ‘aonaosauonis OBLVHSBLNI 170 Table 8. The Signal at the Limit of Detection Integrated Fluorescence, C x l09* axio“5 H 0 Trial cadmium9 _31_ Background -3.18 -2.lO -5.l9 2 -7.65 -l.96 -4. 3 -2.00 -l.13 -4. 4 -2.42 -7.62 -S.19 5 l.l7 3.82 -4.7l Background+ -5.30 -5.59 Mean (signal-background) 3.67 4.52 Standard deviation 0.88 l.lB 0.22 MW. in .(‘lhj' w} m? m: fun,- . 'Dl «1 . w—u—v D . l “Imué— . k .1 (signal-background) * The negative sign comes from a voltage offset in the ADC +Measured before each series of five samples. in the background is much less than the variance in the signals. This is one of the major advantages of the low background nonflame atomizer. This was not the case when a metal vapor lamp was used as a source; background noise was higher in that case. The reason is that the stability of the EDL is much higher than that of the metal vapor discharge lamp. Thus the stray light from the EDL is less noisy, and makes minimal contribution to the total noise present at the limit of detection. The blank (water - background) can be compared to the sample (sample - background) by Student's t test with t defined as t. 3512. I112; (73) S n1+n2 171 where X] and X2 are the mean values for the blank and sample and S is the pooled variance: $2= (74) The experimental value for t (neglecting the variance in the back- ground) is texp = 6.2 The theoretical value of t (8 degrees of freedom) is 5.041 at the 99.9% confidence level. This implies that the difference between 15 g cadMium sample is highly the blank and the signal from a 3 x 10' significant. The true detection limit is obviously less than 3 x l0"15 9. c. Matrix effects. The analytical signal from the graphite braid atomizer was examined as a function of the matrix. The atomizer parameters were varied in an attempt to find the best desolvation, ashing, and atomization temperatures. The porous nature of the GBA should make it possible to find one set of conditions that will produce acceptable results for most matrices. The results are shown in Table 9; they are, however, inconclusive. If selectivity is to be optimized, then a rigorous optimization, as in Chapter IV, must be performed. The GBA should be as free from interferences as are the best furnace atomizers, but this could not be shown. The "optimization" for selectivity consisted of varying only the atomizer temperature; the results, considering the minimum time required for the optimization, were surprisingly good. The very 172. Table 9. Effects of Atomization Temperature on Cadmium in Several Matrices Integrated Fluorescence, C x 107 Voltage Across Atomizer (temperature) Matrix 2.25 V (3.8 sec) 2.45 V (0.4 sec) 3.5 V (0.4 sec) H20 6.86:7% 10.6:3% 2.18:12% serum 3.51:20% 9.39il3% l.77:24% NaCl 2.70:30% 6.4lil3% l.97:l5% Sample 3 pl of l ug/ml Cd Desolvation 0.70 V, 20 sec Ashing l.40 V, 30 sec worst case, Cd in a 1000 ug/ml NaCl, solution showed a depression of 40%. The work of West (46) showed that a To ug/ml NaCl solution depressed the observed AF signal from cadmium by l0%. When a com- mercial furnace was investigated (30), a l000 pg/ml NaCl matrix depressed the AA signal for zinc by 100%. Since zinc normally be- haves similarly to cadmium in both AAS and AFS, a similar depression is to be expected for cadmium. In the light of these investigations, the GBA appears to be superior in selectivity to either the graphite filament or heated graphite furnace. VIII. SUMMARY Nonflame atomization has been studied by two different approaches. First the processes occurring during atomization at a very simple nonflame atomizer have been analyzed. The effects of varying the F3 1 instrumental parameters are elucidated; from a knowledge of how the (I ultimate readout varies with these parameters, the actual events E occurring during atomization can be characterized. The complete charac- E terization of atomization, however, was found to be beyond the scope of this work. If the actual atomization process could be fully Charac- terized, then the analysis conditions could be adjusted to take full advantage of the events occurring during atomization. Although the processes could not be completely elucidated, some qualitative relationships could be discerned. These relationships were then used to guide the construction of a new, second generation nonflame spectrometer. The second generation spectrometer was de- signed for maximum analytical utility. The performance of the new system has been compared to the original system and to the work of other researchers. Nonflame atomization is much easier to study than atomization in a flame. The work performed on the platinum loop atomizer should be expanded in the future. Even though this atomizer has little use in the real world of hard-to-atomize samples, atomization at a heated platinum filament can be studied more easily than atomization at a more complex atomizer. I73 I74 Among the projects that should be studied in detail is the matrix problem. A study of the mechanism of interelement effects would be a valuable contribution to the information known about non- flame atomization. Another interesting study would be to characterize the linearity of the system. The factors influencing linearity are not really known. The nonflame atomizer has great potential in the area of ele- . mental analysis. An instrument employing a flame atomizer, for the a ‘ easily excited elements and a nonflame atomizer for the others could fiji : provide a method for quantitative analysis of all naturally occurring elements other than hydrogen, carbon, nitrogen, oxygen, fluorine, and the noble gases. Certainly the potential justifies more research into nonflame atomization. L I ST 0F REFERENCES u—l o homNOl 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. REFERENCES R. Mavrodineaux and H. Boiteux, "Flame Spectroscopy", John Wiley & Sons, Inc., New York, 1965. J. v. Robinson, "Atomic Absorption", Marcel Dekker, New York, 1966. J. B. Willis, Appl. Optics, L, 1295 (1968). J. D. Winefordner, S. G. Schulman and T. C. O'Haver, "Lumines- cence Spectrometry in Analytical Chemistry", John Wiley & Sons, New York, 1972, p. 132. J. D. Winefordner in "Atomic Absorption Spectroscopy", R. M. Dagnall and T. S. West, Ed., Butterworth & Co., London, 1970, p. 37. B. V. L'vov, Spectrochim. Acta, L;, 761 (1961. Akbar Montaser, Ph. D. Thesis, Michigan State University, 1974. A. Montaser, S. R. Goode and S. R. Crouch, Anal, £339., in press. G. F. Kirkbright, Analyst, 363 609 (1971). J. D. Winefordner and T. J. Vickers, Anal. Chem., 44_(5), 150R (1972). "T L; . Fraunhofer, Ann. Physik (Gilbert's Ann.), 22;.264 (1817). . w. wood, Phil. m" g, 128 (1902). . V. L'vov, J, Engng. Phys., 2, 41 (1959). . I. Belyaev, A. V. Karyakin and A. M. Pchelnisev, 9, Anal. hem. USSR, 25, 852 (1970). . Veillon and M. Margoshes, Spectrochim. Acta, 238, 503 (1968). R B M. Marinkovic and T. J. Vickers, Appl. Spectrosc., 22? 319 (1971). Y C C R. R. Wendt and V. A. Fassel, Anal. Chem., 31, 920 (1965). B. M. Gatehouse and A. Walsh, Spectrochim. Acta,‘lg, 249 (1960). V. G. Mossotti, K. Laqua and N. D. Hagenah, Spectrochim. Acta, 2;. 197 (1967). A. A. Venghiattis, Atomic Absorption Newsletter, 6, 19 (1967). 175 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 176 A. A. King, Astrophys. 0,, 75, 379 (1932). B. V. L'vov and A. D. Khartzyzov, 1h. Prikl. Spektrosk., :4? 413 (1969) B. V. L'vov, Spectrochim. Acta, 248, 53 (1969). H. Massman, Spectrochim. Acta., 238, 215 (1968). H. Massman, ;, Analyt, Chem., 225, 203 (1968). R. Woodriff, R. w. Stone and A. M. Held, Appl. Spectrosc., g, 408 (1968). R. Woodriff, B. R. Culver and K. w. Olson, Appl. Spectrosc.24, 530 (1970). "' R. Woodriff and J. F. Lech, Anal. Chem., 22? 1323 (1972). D. C. Manning and F. F. Fernandez, Atomic Absorption Newsletter 3, 65 (1970). D. Clark, R. M. Dagnall and T. S. West, Anal. Chim. Acta., 229 11 (1973). D. A. Segar and J. G. Gonzalez, Anal. Chem., 22? 7 (1972). C. R. Parker, J. Rowe and D. P. Sandoz, Amer. £39,, §_(8), 53 (1973). " E. Norval and C. R. P. Butler, Anal. Chim. Acta, 38? 47 (1972). J. H. Robinson and P. J. Slevin, Amer. Lab, 4_(8), 10 (1972). J. H. Robinson, 0. K. Nolcott, P. J. Slevin and G. D. Hindman, Anal. Chim. Acta., 22? 13 (1973). J. H. Robinson, G. D. Hindman and P. J. Slevin, Anal. Chim. Acta., 69, 165 (1973). J. B. Headridge and D. R. Smith, Talanta, 19, 833 (1972). K. Fuwa and B. L. Vallee, Anal. Chem., 32; 942 (1963). I. Rubeska and B. Moldan, Appl. Optics, Z: 1341 (1968). . R. Koirtyohann and E. E. Pickett, Anal. Chem., 3;, 601 (1965). S C. Iida, Anal. Lett. 3? 251 (1970). H. T. Delves, Analyst., 223 431 (1970). 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56. 57. 58. 59. 60. 61. 62. 63. 177 A. A. Moncrieff, Archs. Dis. Childh., 39, 1 (1964). Committee Report of the American Academy of Pediatrics, Pedi- atrics, ii? 291 (1969). T. S. West and X. K. Williams, Anal. Chim. Acta.,_4§, 27 (1969). R. G. Anderson, I. S. Maines and T. S. West, Anal. Chim. Acta., ‘51, 355 (1970). J. F. Alder and T. S. West, Anal. Chim. Acta.,_§l, 365 (1970). E? r] J. Aggett and T. S. West, Anal. Chim. Acta., 22, 349 (1971). I r‘fl J. Aggett and T. S. West, Anal. Chim. Acta, 2;, 15 (1971). i x D. Alger, R. G. Anderson, I. S. Maines and T. S. West, Anal. ii. . .EfllE-.AEEE",2l’ 271 (1971). , R. G. Anderson, H. W. Johnson and T. S. West, Anal. Chim. Acta., .57, 281 (1971). L. Ebdon, G. F. Kirkbright and T. S. West, Anal. Chim. Acta., .§§9 39 (1972). J. F. Alder and T. S. West, Anal. Chim. Acta., 58, 331 (1972). K. W. Jackson and T. S. West, Anal. Chim. Acta.,_§g, 182 (1972). L. Ebdon, G. F. Kirkbright and T. S. West, Anal. Chim. Acta., _61, 15 (1972). J. F. Alder, and T. S. West, Anal. Chim. Acta.,_§l, 132 (1972). K. W. Jackson, T. S. West and L. Balchin, Anal. Chim. Acta., _§4, 363 (1973). O. J. Johnson, T. S. West and R. M. Dagnall, Anal. Chim. Acta., :23, 171 (1973). M. D. Amos, Amer._Lgb., g (8), 33 (1970). M. D. Amos, P. A. Bennett, K. G. Brodie, P. W. Y. Lung and J. P. Matousek, Anal. Chem,_4§, 211 (1971). M. P. Bratzel and C. L. Chakrabarti, Anal. Chim. Acta.,_§l, 25 (1972). "“ M. P. Bratzel and C. L. Chakrabarti, Anal. Chim. Acta.,_§3, 1 (1973). ._. R. F. Browner and J. D. Winefordner, Anal. Chem.,_44, 247 (1972). 64. 65. 66. 67. 68. 69. 70. 71. 72. 73. 74. 75. 76. 77. 78. 79. 80. 81. 82. 83. 178 C. J. Molnar, R. D. Reeves, J. D. Winefordner, M. T. Glenn, J. R. Ahlstrom and J. Savory, Appl. §pectrosc., 25, 606 (1972). M. Glenn, J. Savory, L. Hart, T. Glenn and J. D. Winefordner, Anal. Chim. Acta., 52, 263 (1971). ?. M.)Pate1 and J. D. Winefordner, Anal. Chim. Acta., 54, 135 1963 . “‘ F. S. Chuang, B. M. Patel, R. D. Reeves, M. T. Glenn and J. D. Winefordner, 555, 5, Spectr., 15, 6 (1973). 51, B. M. Patel, R. D. Reeves, R. F. Browner, C. J. Molnar and J. D. Winefordner, Appl. Spectrosc., 22, 171 (1973). R. D. Reeves, B. M. Patel, C. J. Molnar and J. D. Winefordner, Anal. Chem., 45, 246 (1973). . waLbfir :35 713mg, R. D. Reeves, C. J. Molnar and J. D. Winefordner, Anal. Chem., 44, 913 (1972). R. Bunsen, Justus Liebigs Annln. Chem., 111, 257 (1859). H. L. Kahn, G. E. Peterson and J. E. Schallis, Atomic Absorption Newsletter, 2, 35 (1968). U. Ulvarson, Acta, Chim. Scand., 21, 641 (1967). H. Brandenburger and H. Bader, Helv. Chim. Acta., 55, 1409 1967 . “' H. Brandenburger, Chimia., 22, 449 (1968). C. Fairless and A. J. Bard., Anal. Lett., 5, 433 (1972). M. P. Bratzel, R. M. Dagnall and J. D. Winefordner, Anal. Chem. Acta., 45, 197 (1969). M. P. Bratzel, R. M. Dagnall and J. D. Winefordner, Appl. Spectrosc., 24, 518 (1970). J. V. Chauvin, M. P. Newton, 0. G. Davis, Anal. Chim. Acta., 55, 291 (1973). H. M. Donega and T. E. Burgess, Anal. Chem., 42, 1521 (1970). J. Y. Hwang, P. A. Ullucci and S. B. Smith, Amer. Leg, 5 (3), 41 (1971). "‘ T. Maruta and T. Takeuchi, Anal. Chim. Acta., 52, 253 (1972). T. Maruta and T. Takeuchi, Anal. Chim. Acta., 55, 5 (1973). 84. 85. 86. 87. 88. 89. 90. 91. 92. 93. 94. 95. 96. 97. 98. 99. 100. 101. 102. 103. 179 J. D. Winefordner, V. Svoboda and L. J. Cline, CRC Crit. Rev. Anal. Chem., 1, 233 (1970). R. M. Dagnall in "Analytical Chemistry - Part 1", University Park Press, Baltimore, MD, 1973, p. 86. B. V. L'vov, "Atomic Absorption Spectrochemical Analysis", American Elsevier, New York, 1970. F. H. Horne and T. G. Anderson,_5. Chem. Phy_,,'55, 2321 (1970). T. J. Walsh and E. A. Hausman in "Treatise on Analytical Chemistry", Pt. II, V. 8, I. M. Kolthoff and P. J. Elving, Ed., John Wiley & Sons, New York, NY, 1963, p. 413. "Photomultiplier Manual", RCA Corporation, Harrison, NJ, 1970. ‘1, "SCR Manual“, General Electric, Syracuse, NY, 1972. J. D. Winefordner and R. A. Staab, Anal. Chem.,_55, 8 165 (1964). "' B. J. Rusel, J. P. Shellon, A. Walsh, Spectrochim. Acta.,_5, 317 (1967). " R. F. Browner, B. M. Patel, T. H. Glenn, M. E. Rietta and J. D. Winefordner, Spectrosc. Lett., 5, 311 (1972). M. B.)Denton and H. V. Malmstadt, Appl. Phys. Lett., 1 , 485 1971 . === L. M. Fraser and J. D. Winefordner, Anal. Chem.,_45, 1693 (1971). V. Malmstadt and C. G. Enke, "Electronics for Scientists", A. Benjamin, NY, 1963, p. 356. D M . Ingle, Jr., and S. R. Crouch, Anal. Chem.,_44, 785 (1972). . Hieftje, Anal. Chem., 44 (6), 81A (1972). >092: M. Montaser and S. R. Crouch, Pittsburgh Conference on Analytical Chemistry and Applied Spectroscopy 1973, Paper No. 119. J. H. Perry, "Chemical Engineer's Handbook", 4th Ed., McGraw- Hill, NY, p. 23-46. . K. Skogerboe in "Flame Emission and Atomic Absorption Spectrometry," . Dean, T. C. Rains, Ed., Marcel Dekker, NY, 1969, p. 385. . A. Jones. I. Langmuir, G. M. J. Mackay, Phys._flgy.,:5g, 201 R J M. K. Murphy, C. Veillon, Anal. Chem., 45, 1468 (1973). H (1927). 104. 105. 106. 107. 108. 109. 110. 111. 112. 113. 114. 115. 116. 117. 118. 119. 120. 121. 180 A. D. Wilson, App]. Optics, 2, 1055 (1963). "Crystal Data", J. D. H. Donnay, Ed., American Crystallographic Association, Brooklyn, NY, 1963 p. 841. R. M. Barnes, University of Massachusetts, Amhearst, MA, personal communication. "Heath Instrumentation for Spectroscopy", Heath/Schlumberger, Benton Harbor, MI, 1970. 2' V; ' fi'n‘J' ~l American Society for Testing and Materials, "Manual on Recom- mended Practices in Spectrophotometry, Third Edition, ASMT, Philadelphia, 1969. A: I Enema“ , O" . . R. 8. Cook and R. Jankow, 5, Chem. Educ., 44, 405 (1972). H. H. )Cary and Arnold 0. Beckman, J. Qp, 555, Amer., 51, 682 1941 "‘ 1“?) T. R. Hogness, F. P. Zscheile, Jr., and A. E. Sidwell, Jr., 5. Phys. Chem.,41, 379 (1937). Leo J. Saidel, A. Robert Goldfarb, and Wallace B. Kalt, Science, 55, 683 (1951). D. D. Tunnicliff, Jr., 5, 92, 555, Amer., 4 , 963 (1955). . Slavin, Anal. Chem., 35 561 (1963). . E. Poulson, Appl. Optics, 5, 99 (1964). . R. Jenkins. Spectrochim. Acta., 255, 567 (1967). . T. Larkins, R. M. Lowe, J. V. Sullivan and A. Walsh, Spectro- W R C. Miranda and P. Conte, Appl. Spectrosc.%25, 557 (1971). D p chim. Acta., 245, 187 (1969). T. J. Vickers, P. J. Sleirn, V. I. Muscat and L. T. Farias, Anal. Chem., 44, 930 (1972). P. M. Beckwith, R. F. Browner and J. D. Winefordner, "High Frequency Excited Electrodeless Sources in Analytical Chemistry," Marcel Dekker, 1974 (In press). P. M. Beckwith, BASF-Wyandotte, Inc., Wyandotte, MI, personal communication. VITA VITA Scott Roy Goode was born on July 9, 1948 in Chicago, Illinois. He attended the Chicago public school system and graduated in 1965 from Mather High School. He then attended the University of Illinois and received a Bachelor of Science degree in Chemistry, in 1969. He pursued undergraduate research under the direction of Professor H.V. Malmstadt while at the University of Illinois. I In 1969 he entered Michigan State University and studied Analytical Chemistry under the guidance of Professor S.R. Crouch. He received his degree in 1974 and commenced employment as an Assistant Professor of Chemistry at the University of South Carolina, Columbia SC.